Cybersecurity researchers have revealed a newly identified attack technique that shows how artificial intelligence chatbots can be manipulated to leak sensitive information with minimal user involvement. The method, known as Reprompt, demonstrates how attackers could extract data from AI assistants such as Microsoft Copilot through a single click on a legitimate-looking link, while bypassing standard enterprise security protections.
According to researchers, the attack requires no malicious software, plugins, or continued interaction. Once a user clicks the link, the attacker can retain control of the chatbot session even if the chat window is closed, allowing information to be quietly transmitted without the user’s awareness.
The issue was disclosed responsibly, and Microsoft has since addressed the vulnerability. The company confirmed that enterprise users of Microsoft 365 Copilot are not affected.
At a technical level, Reprompt relies on a chain of design weaknesses. Attackers first embed instructions into a Copilot web link using a standard query parameter. These instructions are crafted to bypass safeguards that are designed to prevent direct data exposure by exploiting the fact that certain protections apply only to the initial request. From there, the attacker can trigger a continuous exchange between Copilot and an external server, enabling hidden and ongoing data extraction.
In a realistic scenario, a target might receive an email containing what appears to be a legitimate Copilot link. Clicking it would cause Copilot to execute instructions embedded in the URL. The attacker could then repeatedly issue follow-up commands remotely, prompting the chatbot to summarize recently accessed files, infer personal details, or reveal contextual information. Because these later instructions are delivered dynamically, it becomes difficult to determine what data is being accessed by examining the original prompt alone.
Researchers note that this effectively turns Copilot into an invisible channel for data exfiltration, without requiring user-entered prompts, extensions, or system connectors. The underlying issue reflects a broader limitation in large language models: their inability to reliably distinguish between trusted user instructions and commands embedded in untrusted data, enabling indirect prompt injection attacks.
The Reprompt disclosure coincides with the identification of multiple other techniques targeting AI-powered tools. Some attacks exploit chatbot connections to third-party applications, enabling zero-interaction data leaks or long-term persistence by injecting instructions into AI memory. Others abuse confirmation prompts, turning human oversight mechanisms into attack vectors, particularly in development environments.
Researchers have also shown how hidden instructions can be planted in shared documents, calendar invites, or emails to extract corporate data, and how AI browsers can be manipulated to bypass built-in prompt injection defenses. Beyond software, hardware-level risks have been identified, where attackers with server access may infer sensitive information by observing timing patterns in machine learning accelerators.
Additional findings include abuses of trusted AI communication protocols to drain computing resources, trigger hidden tool actions, or inject persistent behavior, as well as spreadsheet-based attacks that generate unsafe formulas capable of exporting user data. In some cases, attackers could manipulate AI development platforms to alter spending controls or leak access credentials, enabling stealthy financial abuse.
Taken together, the research underlines that prompt injection remains a persistent and evolving risk. Experts recommend layered security defenses, limiting AI privileges, and restricting access to sensitive systems. Users are also advised to avoid clicking unsolicited AI-related links and to be cautious about sharing personal or confidential information in chatbot conversations.
As AI systems gain broader access to corporate data and greater autonomy, researchers warn that the potential impact of a single vulnerability increases substantially, underscoring the need for careful deployment, continuous monitoring, and ongoing security research.
When it comes to cybercrime, getting into a system is only half the battle; the real challenge is extracting the stolen data without being detected. Companies often focus on preventing unauthorised access, but they must also ensure that data doesn’t slip out undetected. Hackers, driven by profit, constantly innovate methods to exfiltrate data from corporate networks, making it essential for businesses to understand and defend against these techniques.
The Challenge of Data Exfiltration
Once hackers breach a network, they need to smuggle data out without triggering alarms. Intrusion Detection Systems (IDS) are crucial in this fight. They monitor network traffic and system activities for suspicious patterns that may indicate unauthorised data extraction attempts. IDS can trigger alerts or even automatically block suspicious traffic to prevent data loss. To avoid detection, hackers use obfuscation techniques to disguise their actions. This can involve encrypting data or embedding it within harmless-looking traffic, making it difficult for IDS to identify and block the exfiltration attempts.
Reality vs. Hollywood
In Hollywood movies like "Mission Impossible," data theft is often depicted as a physical heist involving stealth and daring. In reality, hackers prefer remote methods to avoid detection and the risk of getting caught. By exploiting vulnerabilities in web servers, hackers can gain access to a network and search for valuable data. Once they find it, the challenge becomes how to exfiltrate it without triggering security systems.
One common way hackers hide their tracks is through obfuscation. A well-known method of obfuscation is image steganography, where data is embedded within images. This technique allows small amounts of data, such as passwords, to be hidden within images without raising suspicion. However, it is impractical for large datasets due to its low bandwidth and the potential for triggering alarms when numerous images are sent out.
Innovative DNS Data Exfiltration
The Domain Name System (DNS) is essential for internet functionality, translating domain names into IP addresses. Hackers can exploit this by sending data disguised as DNS queries. Typically, corporate firewalls scrutinise unfamiliar DNS requests and block those from untrusted sources. However, a novel method known as "Data Bouncing" has emerged, bypassing these restrictions and making data exfiltration easier for hackers.
How Data Bouncing Works
Data Bouncing leverages trusted web hosts to facilitate DNS resolution. Here’s how it works: hackers send an HTTP request to a reputable domain, like "bbc.co.uk," with a forged "Host" header containing the attacker’s domain. Akami Ghost HTTP servers, configured to resolve such domains, process the request, unknowingly aiding the exfiltration.
Every HTTP request a browser makes to a web server includes some metadata in the request’s headers. One of these header fields is the "Host" field, which specifies the requested domain. Normally, if you request a domain that the IP address doesn’t host, you get an error. However, Akami Ghost HTTP servers are set up to send a DNS request to resolve the domain you’ve asked for, even if it’s outside their network. This means you can send a request to a trusted domain, like "bbc.co.uk," with a "Host" header for "encryptedfilechunk.attackerdomain.com," and the trusted domain carries out the DNS resolution for you.
To prevent data exfiltration, companies need a comprehensive security strategy that includes multiple layers of defence. This makes it harder for hackers to succeed and gives security teams more time to detect and stop them. While preventing intrusions is crucial, detecting and mitigating ongoing exfiltration attempts is equally important to protect valuable data.
As cyber threats take new shapes, so must our defences. Understanding sophisticated exfiltration techniques like Data Bouncing is essential in the fight against cybercrime. By staying informed and vigilant, companies can better protect their data from falling into the wrong hands.
Cybersecurity researchers have identified a wave of attacks targeting outdated versions of the HTTP File Server (HFS) software from Rejetto, aiming to distribute malware and cryptocurrency mining tools. These attacks exploit a critical security flaw known as CVE-2024-23692, which allows hackers to execute arbitrary commands without needing authentication.
CVE-2024-23692 is a high-severity vulnerability discovered by security researcher Arseniy Sharoglazov. It was publicly disclosed in May this year, following a detailed technical report. The flaw is a template injection vulnerability that enables remote attackers to send specially crafted HTTP requests to execute commands on the affected systems. The vulnerability affects HFS versions up to and including 2.3m. In response, Rejetto has issued a warning to users, advising against the use of these versions due to their susceptibility to control by attackers.
Researchers at AhnLab Security Intelligence Center (ASEC) have observed multiple attacks on version 2.3m of HFS. This version remains popular among individuals, small teams, educational institutions, and developers for network file sharing. The attacks likely began after the release of Metasploit modules and proof-of-concept exploits soon after the vulnerability's disclosure.
During these attacks, hackers gather information about the compromised system, install backdoors, and deploy various types of malware. Commands such as "whoami" and "arp" are executed to collect system and user information and identify connected devices. Hackers also add new users to the administrators' group and terminate the HFS process to prevent other threat actors from exploiting the same vulnerability.
In several cases, the XMRig tool, used for mining Monero cryptocurrency, was installed. ASEC researchers attribute one of these attacks to the LemonDuck threat group. Other malware payloads deployed include:
1. XenoRAT: A tool for remote access and control, often used alongside XMRig.
2. Gh0stRAT: Used for remote control and data exfiltration.
3. PlugX: A backdoor associated with Chinese-speaking threat actors, providing persistent access.
4. GoThief: An information stealer that uses Amazon AWS for data exfiltration, capturing screenshots, collecting desktop file information, and sending data to an external command and control server.
AhnLab continues to detect attacks on HFS version 2.3m. Given that the server must be online for file sharing, it remains a lucrative target for hackers. Rejetto recommends users switch to version 0.52.x, which is the latest release despite its lower version number. This version is web-based, requires minimal configuration, and supports HTTPS, dynamic DNS, and administrative panel authentication.
The company has also provided indicators of compromise, including malware hashes, IP addresses of command and control servers, and download URLs for the malware used in these attacks. Users are urged to update their software to the latest version and follow cybersecurity best practices to protect their systems from such vulnerabilities.
By assimilating and addressing these vulnerabilities, users can better secure their systems against these sophisticated attacks.
A newly identified cyber threat group, known as "Unfading Sea Haze," has been secretly infiltrating military and government networks in the South China Sea region since 2018, according to a recent report by Bitdefender researchers. The group's activities align with Chinese geopolitical interests, focusing on gathering intelligence and conducting espionage. Unfading Sea Haze shares many tactics, techniques, and procedures (TTPs) with other Chinese state-sponsored hacking groups, particularly APT41.
The group's attacks typically begin with spear-phishing emails containing malicious ZIP files disguised as legitimate documents. These ZIP files, often named to appear as Windows Defender installers, contain LNK files with obfuscated PowerShell commands. If an ESET security executable is detected on the target system, the attack is halted. Otherwise, the PowerShell script uses Microsoft's msbuild.exe to launch fileless malware directly into memory, leaving no traces on the victim's machine.
The code executed by MSBuild installs a backdoor called 'SerialPktdoor,' which gives the attackers remote control over the compromised system. Additionally, the hackers use scheduled tasks and manipulate local administrator accounts to maintain their presence on the network. By resetting and enabling the typically disabled local admin account, they create a hidden profile for continuous access.
Unfading Sea Haze employs a variety of custom tools and malware. Among these are 'xkeylog,' a keylogger for capturing keystrokes, info-stealers targeting browser data, and PowerShell scripts for extracting information. Since 2023, the group has adopted stealthier methods, such as abusing msbuild.exe to load C# payloads from remote SMB shares and deploying different variants of the Gh0stRAT malware.
Bitdefender has identified several Gh0stRAT variants used by the hackers:
1. SilentGh0st: A variant with extensive functionality through numerous commands and modules.
2. InsidiousGh0st: A Go-based evolution with enhanced capabilities, including TCP proxy, SOCKS5, and improved PowerShell integration.
3. TranslucentGh0st, EtherealGh0st, and FluffyGh0st: Newer variants designed for evasive operations with dynamic plugin loading and a lighter footprint.
Earlier attacks utilised tools like Ps2dllLoader for loading .NET or PowerShell code into memory and SharpJSHandler, a web shell for executing encoded JavaScript via HTTP requests. The group also created a tool to monitor newly connected USB and Windows Portable Devices every ten seconds, reporting device details and specific files to the attackers.
For data exfiltration, Unfading Sea Haze initially used a custom tool named 'DustyExfilTool,' which securely extracted data via TLS over TCP. In more recent attacks, the group has shifted to using a curl utility and the FTP protocol, with dynamically generated credentials that are frequently changed to enhance security.
The sophisticated techniques employed by Unfading Sea Haze highlight the need for robust cybersecurity defences. Organisations should implement a comprehensive security strategy that includes regular patch management, multi-factor authentication (MFA), network segmentation, traffic monitoring, and advanced detection and response tools.
By adopting these measures, organisations can better defend against the persistent and evolving threats posed by groups like Unfading Sea Haze. The group's ability to remain undetected for six years sets a strong precedent for the critical importance of vigilance and continuous improvement in cybersecurity practices.
According to a recent discovery by Varonis Threat Labs, two new techniques have emerged that pose a significant threat to data security within SharePoint, a widely used platform for file management. These techniques enable users to evade detection and retreat files without triggering alarm bells in audit logs.
Technique 1: Open in App Method
The first technique leverages SharePoint's "open in app" feature, allowing users to access and download files while leaving behind only access events in the file's audit log. This method, which can be executed manually or through automated scripts, enables rapid exfiltration of multiple files without raising suspicion.
Technique 2: SkyDriveSync User-Agent
The second technique exploits the User-Agent for Microsoft SkyDriveSync, disguising file downloads as sync events rather than standard downloads. By mislabeling events, threat actors can bypass detection tools and policies, making their activity harder to track.
Implications for Security
These techniques pose a significant challenge to traditional security tools such as cloud access security brokers and data loss prevention systems. By hiding downloads as less suspicious access and sync events, threat actors can circumvent detection measures and potentially exfiltrate sensitive data unnoticed.
Microsoft's Response
Despite Varonis disclosing these methods to Microsoft, the tech giant has designated them as a "moderate" security concern and has not taken immediate action to address them. As a result, these vulnerabilities remain in SharePoint deployments, leaving organisations vulnerable to exploitation.
Recommendations for Organisations
To alleviate the risk posed by these techniques, organisations are advised to closely monitor access events in their SharePoint and OneDrive audit logs. Varonis recommends leveraging User and Entity Behavior Analytics (UEBA) and AI features to detect and stop suspicious activities, such as mass file access.
What Are the Risks?
While SharePoint and OneDrive are essential tools for facilitating file access in organisations, misconfigured permissions and access controls can inadvertently expose sensitive data to unauthorised users. Threat actors often exploit these misconfigurations to exfiltrate data, posing a significant risk to organisations across various industries.
Detection and Prevention Strategies
To detect and prevent unauthorised data exfiltration, organisations should implement detection rules that consider behavioural patterns, including frequency and volume of sync activity, unusual device usage, and synchronisation of sensitive folders. By analysing these parameters, organisations can identify and mitigate potential threats before they escalate.