Search This Blog
Powered by Blogger.

Blog Archive

Labels

Footer About

Footer About

Labels

Load More
Trendy Right Now

Popular Posts

Latest News

CrossCurve Bridge Hit by $3 Million Exploit after Smart Contract Flaw

CrossCurve, a cross-chain bridge formerly known as EYWA, has suffered a major cyberattack after hackers exploited a vulnerability in its sma...

All the recent news you need to know
Windows
Audit ClawHub Cyber Attacks macOS Windows

Experts Find Malicious ClawHub Skills Stealing Data from OpenClaw


Koi Security’s security audit of 2,857 skills on ClawHub found 341 malicious skills via multiple campaigns. Users are exposed to new supply chain threats. 

ClawHub is a marketplace made to help OpenClaw users in finding and installing third-party skills. It is a part of the OpenClaw project, a self-hosted artificial intelligence (AI) assistant aka Moltbot and Clawdbot. 

Koi Security's analysis with OpenClaw bot “Alex” revealed that 335 skills use malicious pre-requisite to install an Apple macOS stealer called (Atomic Stealer). The activity goes by the code name ClawHavoc. 

According to Koi research Oren Yomtov, "You install what looks like a legitimate skill – maybe solana-wallet-tracker or youtube-summarize-pro. The skill's documentation looks professional. But there's a 'Prerequisites' section that says you need to install something first.”

Instruction steps:

Windows users are asked to download file “openclaw-agent.zip” from a GitHub repository.

macOS users are asked to copy an installation script hosted at glot[.]io and paste it in the Terminal application. 

Threat actors are targeting macOS users because of an increase in purchase of Mac Minus to use the AI assistant 24x7. 

In the password-protected archive, the trojan has keylogging functionality to steal credentials, API keys, and other important data on the device. Besides this, the glot[.]io script includes hidden shell commands to retrieve next-stage payloads from a threat-actor controlled infrastructure. 

This results in getting another IP address ("91.92.242[.]30") to get another shell script, which is modified to address the same server to get a universal Mach-O binary that shows traits persistent with Atomic Stealer, a commodity stealer that threat actors can buy for $500-1000/month that can extract data from macOS hosts.

The issue is that anyone can post abilities to ClawHub because it is open by default. At this point, the only requirement is that a publisher have a GitHub account that is at least a week old. 

Peter Steinberger, the founder of OpenClaw, is aware of the problem with malicious abilities and has subsequently implemented a reporting option that enables users who are signed in to report a skill. According to the documentation, "Each user can have up to 20 active reports at a time," "Skills with more than 3 unique reports are auto-hidden by default.”


Read more »
Malware Campaigns
AI Prompt AI Security AI security risks AI technology Cyber Attacks LLM LLM cybersecurity Malware Campaigns

Promptware Threats Turn LLM Attacks Into Multi-Stage Malware Campaigns

 

Large language models are now embedded in everyday workplace tasks, powering automated support tools and autonomous assistants that manage calendars, write code, and handle financial actions. As these systems expand in capability and adoption, they also introduce new security weaknesses. Experts warn that threats against LLMs have evolved beyond simple prompt tricks and now resemble coordinated cyberattacks, carried out in structured stages much like traditional malware campaigns. 

This growing threat category is known as “promptware,” referring to malicious activity designed to exploit vulnerabilities in LLM-based applications. It differs from basic prompt injection, which researchers describe as only one part of a broader and more serious risk. Promptware follows a deliberate sequence: attackers gain entry using deceptive prompts, bypass safety controls to increase privileges, establish persistence, and then spread across connected services before completing their objectives.  

Because this approach mirrors conventional malware operations, long-established cybersecurity strategies can still help defend AI environments. Rather than treating LLM attacks as isolated incidents, organizations are being urged to view them as multi-phase campaigns with multiple points where defenses can interrupt progress.  

Researchers Ben Nassi, Bruce Schneier, and Oleg Brodt—affiliated with Tel Aviv University, Harvard Kennedy School, and Ben-Gurion University—argue that common assumptions about LLM misuse are outdated. They propose a five-phase model that frames promptware as a staged process unfolding over time, where each step enables the next. What may appear as sudden disruption is often the result of hidden progress through earlier phases. 

The first stage involves initial access, where malicious prompts enter through crafted user inputs or poisoned documents retrieved by the system. The next stage expands attacker control through jailbreak techniques that override alignment safeguards. These methods can include obfuscated wording, role-play scenarios, or reusable malicious suffixes that work across different model versions. 

Once inside, persistence becomes especially dangerous. Unlike traditional malware, which often relies on scheduled tasks or system changes, promptware embeds itself in the data sources LLM tools rely on. It can hide payloads in shared repositories such as email threads or corporate databases, reactivating when similar content is retrieved later. An even more serious form targets an agent’s memory directly, ensuring malicious instructions execute repeatedly without reinfection. 

The Morris II worm illustrates how these attacks can spread. Using LLM-based email assistants, it replicated by forcing the system to insert malicious content into outgoing messages. When recipients’ assistants processed the infected messages, the payload triggered again, enabling rapid and unnoticed propagation. Experts also highlight command-and-control methods that allow attackers to update payloads dynamically by embedding instructions that fetch commands from remote sources. 

These threats are no longer theoretical, with promptware already enabling data theft, fraud, device manipulation, phishing, and unauthorized financial transactions—making AI security an urgent issue for organizations.
Read more »
Tribal Clinics
Data Breach Medical Data Leak Ransomware attack Rhysida Ransomware Tribal Clinics

Rhysida Ransomware Hits California Tribal Clinics, Leaks SSNs and Medical Data

 

A recent ransomware attack has disrupted healthcare services and exposed sensitive patient data at the MACT Health Board, which operates clinics serving American Indian communities in California’s Sierra Foothills. The cybercriminal group Rhysida has claimed responsibility for the November 2025 breach and has listed MACT on its data leak site, demanding a ransom of eight bitcoin, valued at about 662,000 dollars at the time. Although MACT has notified affected patients, the organization has not confirmed Rhysida’s claims or disclosed how many individuals were impacted.

According to MACT’s notice to victims, an unauthorized party accessed some files on its systems between November 12 and November 20, 2025, leading to serious exposure of personal and medical information. Compromised data includes names, Social Security numbers, and detailed medical information such as diagnoses, doctors, insurance details, medications, test results, images, and records of care and treatment. In response, MACT is offering eligible victims free identity monitoring, recognizing the heightened risk of identity theft and fraud.

The attack caused significant operational disruption across MACT’s clinics starting November 20, 2025, affecting phone services, prescription ordering, and appointment scheduling. Phone lines were restored by December 1, but some specialized imaging services were still offline as of January 22, illustrating the long-term impact such incidents can have on patient care. The Board declined to answer detailed questions about the breach, including whether a ransom was paid or how the attackers infiltrated the network.

Rhysida, which emerged in May 2023, runs a ransomware-as-a-service model, providing its malware and infrastructure to affiliates who carry out attacks. Its ransomware both steals data and encrypts systems, with victims pressured to pay for deletion of stolen information and for decryption keys. The group has claimed responsibility for 102 confirmed attacks and an additional 157 unacknowledged incidents, with an average ransom demand of around 884,000 dollars. At least 24 of its confirmed attacks have targeted healthcare entities, compromising about 3.83 million records, including high-profile breaches at MedStar Health, Spindletop Center, and Cytek Biosciences.

The MACT incident highlights a broader surge in ransomware targeting US healthcare providers. Comparitech researchers documented 109 confirmed ransomware attacks against hospitals, clinics, and other care providers in 2025 alone, affecting nearly 8.9 million records. These attacks can force organizations back to pen-and-paper operations, trigger appointment cancellations, and even require patient diversions, putting both safety and privacy at risk. MACT, which serves five California counties—Mariposa, Amador, Alpine, Calaveras, and Tuolumne—through about a dozen clinics offering medical, dental, behavioral, optometry, and chiropractic care, now faces the dual challenge of restoring services and rebuilding trust with its community.
Read more »
WPLMS Vulnerability
CVE-2024-10470 malware injection Search Engine Cloaking SEO Spam Attack Theme Exploitation Vulnerabilities and Exploits Website Compromise Wordpress Security WPLMS Vulnerability

Security Analysts Warn of Shadow Directory Techniques Targeting WordPress


 

Several theme-level vulnerabilities coupled with evolving abuse tactics are demonstrating once again how vulnerable WordPress becomes when multiple vulnerabilities are aligned. An unauthenticated file access and deletion vulnerability has been disclosed in the WPLMS theme-tracked as CVE-2024-10470 and assigned a CVSS score of 9.8-which has exposed thousands of learning management deployments. 

A significant risk exists as a result of the issue in more than 28,000 active installations, which enables attackers to read or remove sensitive files such as wp-config.php, thereby lowering the barrier to full site compromise, data exposure, and operational disruption. 

Not only does the vulnerability itself pose a serious threat, but its intersection with a broader wave of hostile activity that has already targeted WordPress ecosystems at a significant scale makes this threat particularly acute. This is in keeping with recent research by Sucuri that shows threat actors are utilizing malicious JavaScript injections to weaponize distributed brute-force campaigns against compromised sites. 

Instead of attacking targets directly, injected code quietly conscripts unsuspecting web browsers, creating a distributed attack platform based on normal web traffic. Earlier campaigns were focused on crypto drainers and Web3 phishing redirects, but the latest iteration, which has been observed on over 700 websites, uses leaked and commonly used credentials to systematically brute-force additional WordPress installations, representing a dramatic shift from these earlier campaigns. 

This development demonstrates how critical theme vulnerabilities and indirect attack mechanisms are being combined to amplify impact, accelerate lateral spread, and undermine trust in compromised WordPress environments by chaining together critical theme vulnerabilities and indirect attack mechanisms.

It was discovered that the attackers had manipulated the way pages were rendered to specific visitors, rather than altering visible site content. During search engine results, valid titles and descriptions were replaced with casino- and gambling-related text, suggesting that this was done deliberately to poison the indexing of search engines. 

Notably, the spam did not appear on highly trafficked areas, such as the homepage or blog posts. In place of static, low-maintenance pages, it appeared on static, low-maintenance pages like About Us, Contact Us, Privacy Policy, and Terms & Conditions—sections that are generally static and rarely examined. Consequently, the placement alone indicates a deliberate attempt to avoid detection while maximising SEO abuse. 

Verification of the affected pages through the WordPress administrative interface revealed no signs of compromise. They retained their expected permalinks and the content displayed within the editor was entirely valid. Based on a direct review of the underlying database, including the records in the wp_posts table, it was determined that no unauthorized modifications had occurred to the content storage system. 

In the present state, there is little to explain why search engines index spam, while human visitors only see benign content, since neither the page source nor the database indicate any anomalies. A turning point occurred when investigators altered the browser’s User-Agent string to emulate a crawler’s behavior.

Consequently, the same URLs exhibited entirely different content when viewed under these conditions. Instead of displaying standard corporate or legal information, pages previously showing standard corporate or legal information appeared full-screen casino spam, confirming that cloaking has been implemented. 

By using this conditional delivery mechanism, attackers were able to target search engine bots selectively and conceal the malicious payload from site administrators and regular users. A deeper technical review indicated that the compromise extends beyond simple content injection and that it involves a more advanced method for manipulating the permalinks and page rendering logic within WordPress itself without leaving conventional forensic traces. 

As a result of a deeper technical analysis, the root cause was determined to be an envato-setup-export.php component of the WPLMS theme which lacked critical security controls. In this code, the zip_file parameter is not validated or sanitized, which enables an attacker to reference any file located on the server arbitrarily.

Once a target file is selected, the application logic proceeds to read its contents using the readfile() function before deleting it using the unlink() function. By executing this unsafe sequence, unauthenticated attackers are effectively capable of removing critical configuration files, including WordPress' configuration file, user-generated content, and other important assets from the site, thereby significantly weakening the site's security posture. This vulnerability has a broader impact than deleting isolated files. 

Upon removing the WordPress config.php file, WordPress will be placed in an installation state, as it will be disconnected from the existing database. By exploiting the setup process, an attacker can link the site with a database under his/her control, allowing full administrative control of the website. 

A file-handling vulnerability that initially appears to be a vulnerability is transformed into a complete site compromise that can result in data exfiltration, persistent backdoors, and abuse of trust among users. As the theme is widely deployed across multiple sectors, the vulnerability is a high-value target for both opportunistic and coordinated attacks, thereby amplifying the threat. 

Initially disclosed by security researcher Foxyyyy through Wordfence's Bug Bounty Program, Foxyyyy received a $900 bounty. Following this publication, Wordfence has issued a recommendation for all WPLMS users to upgrade to version 4.963 or later, which contains a fix for the vulnerability known as CVE-2024-10470. Because the flaw is global in scope and severe, administrators are encouraged to prioritize patching. 

As part of enhancing baseline defenses, security teams should harden servers and applications, maintain offline backups, enforce strong authentication practices, such as unique passwords and multi-factor authentication, and make sure that operating systems, plugins, and security signatures are fully updated. 

When similar vulnerabilities arise in the future, these measures are critical to reducing exploitation risk and limiting the blast radius. The malicious functionality was found to be completely hidden within an index.php file placed in fabricated directories which reflected legitimate WordPress permalinks as a result of the subsequent investigation. 

A traffic gatekeeper file serves as a dynamic gatekeeper, determining whether to serve authentic-looking content or spam payloads based on the nature of incoming requests. As a result of conditional logic, routine visits by site owners and regular users did not raise any immediate suspicion, while search engine crawlers were selectively given content that was designed to be indexed. 

A bot-detection routine, which was embedded at the heart of this mechanism, was designed to efficiently identify search engine traffic with an extremely high level of precision. The attackers specifically targeted multiple Google-related User-Agent strings in order to effectively differentiate search engine crawlers from human traffic. 

By identifying these identifiers in a request, the malware bypasses the legitimate page altogether and loads a secondary file directly into the browser, readme.txt, rendering its contents directly in the browser. 

While this file appeared harmless, it contained a complete HTML document containing spam content specifically designed to influence the rankings of search engines. In order to conceal the compromise, the malicious script instead included indexx.php, an HTML static capture of the original page for non-bot traffic. This approach portrayed the site as entirely normal to administrators and users, effectively cloaking the compromise. 

In addition, the readme.txt payload further emphasized the sophisticated nature of the campaign by providing deceptive information. A file containing more than 600 lines of HTML, JavaScript, and stolen styling elements impersonated a high authority online retail site spanning more than 600 lines. 

In addition to reusing CSS, metadata, and structural elements related to Etsy, the attackers attempted to add credibility to the webpage by using underlying content that promoted Indonesian gambling schemes referred to as "Slot Gacor." The attackers also inserted extensive Schema Markup, using JSON-LD, to enhance visibility in search results. 

In addition to creating fabricated product listings, aggregate ratings, and pricing information, the attackers effectively manipulated how search engine snippets portrayed the pages. Consequently, what was in fact a cloaked spam page could appear in search results presenting itself as an attractive, highly rated product, leading to an increase in click-through rates and an extension of infection lifespan. 

Instead of treating symptoms individually, remediation efforts focused on fully removing the malicious infrastructure. As a result of identifying and deleting all unauthorized directories that replicated WordPress permalinks, as well as the associated malicious files, the site was then tested using multiple User-Agent configurations, ensuring that content delivery was consistent and no longer dependent on the User-Agent configuration. 

A complete review of file permissions, resetting of all administrative, hosting, FTP, and database credentials was performed in parallel with the site owner's advice to request a fresh index from search engines upon completion of the cleanup. 

In addition, security specialists recommended that an extensive post-incident hardening process be implemented, including the removal of unknown user accounts, checking core WordPress files such as index.php and .htaccess, and regularly monitoring server logs for suspicious outbound connections. 

As part of the recommendations, additional safeguards were emphasized, including maintaining offline backups, updating all components, scanning administrator devices for malware, and deploying a Web Application Firewall to prevent cloaking-based campaigns from taking hold in the future, and reducing the likelihood of reinfection. 

According to the findings of subsequent investigation, the malicious functionality was completely contained within an index.php file located in fabricated directories that mimicked legitimate WordPress permalinks. By dynamically analyzing the nature of the incoming request, this file determined whether to serve authentic-looking page content or spam payload. 

Using conditional logic, routine visits by site owners and regular users did not raise suspicion, while search engine crawlers were given manipulated content intended for indexing selectively. As a key component of this methodology, a bot-detection routine was developed which was capable of identifying search engine traffic with high precision. 

As a result of the attackers' explicit targeting of multiple Google-related User-Agent strings, they have been able to identify crawlers from human users reliably. In response to a request matching these identifiers, the malware bypassed the legitimate site entirely and loaded a secondary file, readme.txt, which was displayed directly in the browser. 

However, the file contained an entire HTML document with spam content designed to influence search engine rankings despite its seemingly innocent name. In order to conceal the compromise, the malicious script instead included indexx.php, an HTML static capture of the original page for non-bot traffic. This approach portrayed the site as entirely normal to administrators and users, effectively cloaking the compromise.

In addition, the readme.txt payload further emphasized the sophisticated nature of the campaign by providing deceptive information. A file containing more than 600 lines of HTML, JavaScript, and stolen styling elements impersonated a high authority online retail site spanning more than 600 lines. 

In addition to reusing CSS, metadata, and structural elements related to Etsy, the attackers attempted to add credibility to the webpage by using underlying content that promoted Indonesian gambling schemes referred to as "Slot Gacor." The attackers also inserted extensive Schema Markup, using JSON-LD, to enhance visibility in search results. 

In addition to creating fabricated product listings, aggregate ratings, and pricing information, the attackers effectively manipulated how search engine snippets portrayed the pages. Due to this, what was in fact a cloaked spam page appeared in search results as a product that was perceived as legitimate and highly rated, resulting in increased click-throughs and prolonged infection.

In contrast to treating symptoms in isolation, remediation efforts focused on eliminating the malicious infrastructure completely. A comprehensive investigation of all unauthorized directories replicating WordPress permalinks was conducted, together with the associated malicious files, and all unauthorized directories were deleted. 

Following the verification of consistent content delivery, multiple User-Agent configurations were used to verify that any conditionality was eliminated. In parallel, file permissions were reviewed, all administrative, hosting, FTP, and database credentials were reset, and the website owner was instructed to request that search engines refresh their index once the cleanup had been completed.

Additionally, security specialists recommend that all WordPress user accounts be removed, core WordPress files such as index.php and .htaccess be verified, and server logs for suspicious outbound connections be monitored regularly. 

It is important to maintain offline backups, keep all components updated, scan administrator devices for malware, and deploy a Web Application Firewall as additional safeguards so that reinfection will be reduced and similar cloaking-based campaigns will not occur again. 

Overall, the investigation indicates how a number of weaknesses were purposefully chained together to create a highly resilient and covert compromise. This attack did not utilize overt defacement or obvious content injection, but rather exploited a critical WordPress theme vulnerability combined with cloaking techniques to manipulate WordPress' rendering of content in accordance with the visitor's identity. 

Using CVE-2024-10470 in the WPLMS theme, they were able to remove or access sensitive files, weakening the security of the site and creating conditions for deeper manipulation of data without altering it. Several recent Sucuri findings confirm the shift towards indirect abuse mechanisms in WordPress-focused attacks, which correspond with the broader campaign. 

A malicious JavaScript injection was used by threat actors to manipulate unsuspecting visitors' browsers, effectively converting legitimate traffic into a distributed attack network via JavaScript injections. The most recent phase of this activity was observed across over 700 compromised sites.

It utilised leaked and commonly used credentials to brute-force additional WordPress installations, despite earlier iterations focusing on crypto drainers and Web3 phishing redirections. This evolution shows a deliberate movement toward techniques that are scalable and low-noise, thus maximizing reach and minimizing immediate detection.

An extremely sophisticated cloaking component was implemented in the attack Malicious logic was incorporated into index.php files located in fabricated directories that were intended to mirror legitimate WordPress permalinks, providing access to legitimate permalinks as well as identifying search engine crawlers using multiple Google-specific User-Agent strings to serve as traffic gatekeepers. 

Upon identifying crawler traffic, an additional payload was served from a file deceptively titled readme.txt which contained a fully formatted HTML document that was optimized for search engine indexing. However, human visitors and site administrators received indexx.php, a static HTML snapshot of the legitimate page, thereby ensuring that the site appeared normal during routine inspections. 

Over 600 lines of code were incorporated into the spam payload in an effort to fool automated systems. Combining HTML, JavaScript, and stolen design elements created the illusion of an e-commerce platform with high authority. 

A number of technical components associated with Etsy, including CSS, metadata, and structural components, were reused by the attackers, together with extensive JSON-LD schema markup that contained fabricated product data, ratings, and pricing. As a result, the attackers manipulated how search engines interpreted and displayed the pages. 

The result was the presentation of cloaked gambling content-promoting Indonesian "Slot Gacor" schemes-as seemingly legitimate, highly rated listings in search results, increasing click-through rates and extending the lifespan of the infection. 

Rather than addressing individual symptoms, remediation efforts focused on dismantling the attack infrastructure as a whole. A thorough examination was conducted to identify and remove all unauthorized directories that replicated WordPress permalinks along with the malicious files associated with them.

Tests were conducted using a variety of user agents to confirm the removal of conditional content delivery. Additionally, site owners were advised to request re-indexing by search engines following cleanup by reviewing file permissions, resetting all administrative, hosting, FTP, and database credentials, as well as re-adjusting all administrative credentials.

The security specialists reiterated that post-incident hardening is extremely important in order to prevent recurrences. It was recommended that unknown user accounts be audited and removed, core WordPress files such as index.php and .htaccess be verified for integrity, server logs for suspicious outbound connections be monitored, offline backups be maintained, all components are kept up to date, malware is scanned on administrator systems, and a web application firewall be implemented. 

Collectively, these steps play a critical role in reducing the risk of reinfection and defending against future cloaking-related campaigns that increasingly blur the boundaries between content abuse, SEO manipulation, and the compromise of the entire site.
Read more »
Olympics
AI threats Cyber Security Cyber Threats Data Privacy Olympics

Italy Steps Up Cyber Defenses as Milano–Cortina Winter Olympics Approach

 



Inside a government building in Rome, located opposite the ancient Aurelian Walls, dozens of cybersecurity professionals have been carrying out continuous monitoring operations for nearly a year. Their work focuses on tracking suspicious discussions and coordination activity taking place across hidden corners of the internet, including underground criminal forums and dark web marketplaces. This monitoring effort forms a core part of Italy’s preparations to protect the Milano–Cortina Winter Olympic Games from cyberattacks.

The responsibility for securing the digital environment of the Games lies with Italy’s National Cybersecurity Agency, an institution formed in 2021 to centralize the country’s cyber defense strategy. The upcoming Winter Olympics represent the agency’s first large-scale international operational test. Officials view the event as a likely target for cyber threats because the Olympics attract intense global attention. Such visibility can draw a wide spectrum of malicious actors, ranging from small-scale cybercriminal groups seeking disruption or financial gain to advanced threat groups believed to have links with state interests. These actors may attempt to use the event as a platform to make political statements, associate attacks with ideological causes, or exploit broader geopolitical tensions.

The Milano–Cortina Winter Games will run from February 6 to February 22 and will be hosted across multiple Alpine regions for the first time in Olympic history. This multi-location format introduces additional security and coordination challenges. Each venue relies on interconnected digital systems, including communications networks, event management platforms, broadcasting infrastructure, and logistics systems. Securing a geographically distributed digital environment exponentially increases the complexity of monitoring, response coordination, and incident containment.

Officials estimate that the Games will reach approximately three billion viewers globally, alongside around 1.5 million ticket-holding spectators on site. This scale creates a vast digital footprint. High-visibility services, such as live streaming platforms, official event websites, and ticket purchasing systems, are considered particularly attractive targets. Disrupting these services can generate widespread media attention, cause public confusion, and undermine confidence in the organizers’ ability to safeguard critical digital operations.

Italy’s planning has been shaped by recent Olympic experience. During the 2024 Paris Summer Olympics, authorities recorded more than 140 cyber incidents. In 22 cases, attackers managed to gain access to information systems. While none of these incidents disrupted the competitions themselves, the sheer volume of hostile activity demonstrated the persistent pressure faced by host nations. On the day of the opening ceremony in Paris, France’s TGV high-speed rail network was also targeted in coordinated physical sabotage attacks involving explosive devices. This incident illustrated how large global events can attract both cyber threats and physical security risks at the same time.

Italian cybersecurity officials anticipate comparable levels of hostile activity during the Milano–Cortina Games, with an additional layer of complexity introduced by artificial intelligence. AI tools can be used by attackers to automate technical tasks, enhance reconnaissance, and support more convincing phishing and impersonation campaigns. These techniques can increase the speed and scale of cyber operations while making malicious activity harder to detect. Although authorities currently report no specific, elevated threat level, they acknowledge that the overall risk environment is becoming more complex due to the growing availability of AI-assisted tools.

The National Cybersecurity Agency’s defensive approach emphasizes early detection rather than reactive response. Analysts continuously monitor open websites, underground criminal communities, and social media channels to identify emerging threat patterns before they develop into direct intrusion attempts. This method is designed to provide early warning, allowing technical teams to strengthen defenses before attackers move from planning to execution.

Operational coordination will involve multiple teams. Around 20 specialists from the agency’s operational staff will focus exclusively on Olympic-related cyber intelligence from the headquarters in Rome. An additional 10 senior experts will be deployed to Milan starting on February 4 to support the Technology Operations Centre, which oversees the digital systems supporting the Games. These government teams will operate alongside nearly 100 specialists from Deloitte and approximately 300 personnel from the local organizing committee and technology partners. Together, these groups will manage cybersecurity monitoring, incident response, and system resilience across all Olympic venues.

If threats keep developing during the Games, the agency will continuously feed intelligence into technical operations teams to support rapid decision-making. The guiding objective remains consistent. Detect emerging risks early, interpret threat signals accurately, and respond quickly and effectively when specific dangers become visible. This approach reflects Italy’s broader strategy to protect the digital infrastructure that underpins one of the world’s most prominent international sporting events.


Read more »
Technology
AI Risks Open-Source AI Models Technology

Open-Source AI Models Pose Growing Security Risks, Researchers Warn

Hackers and other criminals can easily hijack computers running open-source large language models and use them for illicit activity, bypassing the safeguards built into major artificial intelligence platforms, researchers said on Thursday. The findings are based on a 293-day study conducted jointly by SentinelOne and Censys, and shared exclusively with Reuters. 

The research examined thousands of publicly accessible deployments of open-source LLMs and highlighted a broad range of potentially abusive use cases. According to the researchers, compromised systems could be directed to generate spam, phishing content, or disinformation while evading the security controls enforced by large AI providers. 

The deployments were also linked to activity involving hacking, hate speech, harassment, violent or graphic content, personal data theft, scams, fraud, and in some cases, child sexual abuse material. While thousands of open-source LLM variants are available, a significant share of internet-accessible deployments were based on Meta’s Llama models, Google DeepMind’s Gemma, and other widely used systems, the researchers said. 

They identified hundreds of instances in which safety guardrails had been deliberately removed. “AI industry conversations about security controls are ignoring this kind of surplus capacity that is clearly being utilized for all kinds of different stuff, some of it legitimate, some obviously criminal,” said Juan Andres Guerrero-Saade, executive director for intelligence and security research at SentinelOne. He compared the problem to an iceberg that remains largely unaccounted for across the industry and the open-source community. 

The study focused on models deployed using Ollama, a tool that allows users to run their own versions of large language models. Researchers were able to observe system prompts in about a quarter of the deployments analyzed and found that 7.5 percent of those prompts could potentially enable harmful behavior. 

Geographically, around 30 per cent of the observed hosts were located in China, with about 20 per cent based in the United States, the researchers said. Rachel Adams, chief executive of the Global Centre on AI Governance, said responsibility for downstream misuse becomes shared once open models are released.  “Labs are not responsible for every downstream misuse, but they retain an important duty of care to anticipate foreseeable harms, document risks, and provide mitigation tooling and guidance,” Adams said.  

A Meta spokesperson declined to comment on developer responsibility for downstream abuse but pointed to the company’s Llama Protection tools and Responsible Use Guide. Microsoft AI Red Team Lead Ram Shankar Siva Kumar said Microsoft believes open-source models play an important role but acknowledged the risks. 

“We are clear-eyed that open models, like all transformative technologies, can be misused by adversaries if released without appropriate safeguards,” he said. 

Microsoft conducts pre-release evaluations and monitors for emerging misuse patterns, Kumar added, noting that “responsible open innovation requires shared commitment across creators, deployers, researchers, and security teams.” 

Ollama, Google and Anthropic did not comment. 
Read more »
Subscribe to: Comments (Atom)

Featured