VulnHub

Real-time Cybersecurity News

Top AI Agents Built to Catch Malicious Code Can Be Tricked Into Running It

The Hacker News
Vulnerability

Overview

Researchers from the AI Now Institute have discovered a significant flaw in AI coding agents, specifically Anthropic's Claude Code and OpenAI's Codex. Their proof-of-concept, termed 'Friendly Fire', reveals that these AI systems can inadvertently execute malicious code when tasked with scanning open-source projects for vulnerabilities. This occurs when the AI operates in an autonomous mode that allows it to approve and run code without human oversight. The implications of this are serious, as it could lead to unintentional security breaches on users' machines, potentially exposing sensitive information or allowing further attacks. Developers and organizations using these AI tools need to be aware of this risk and implement safeguards to prevent such incidents.

Key Takeaways

  • Affected Systems: Anthropic's Claude Code, OpenAI's Codex
  • Action Required: Users should ensure AI coding agents operate under stringent oversight and avoid enabling autonomous execution of code.
  • Timeline: Disclosed on October 25, 2023

Original Article Summary

Ask an AI coding agent to scan open-source code for security holes, and it might run the attacker's code on your own machine instead. That is the finding in a proof-of-concept published Wednesday by the AI Now Institute, an attack it calls "Friendly Fire." It works against Anthropic's Claude Code and OpenAI's Codex when either is running in an autonomous mode that approves its own

Impact

Anthropic's Claude Code, OpenAI's Codex

Exploitation Status

No active exploitation has been reported at this time. However, organizations should still apply patches promptly as proof-of-concept code may exist.

Timeline

Disclosed on October 25, 2023

Remediation

Users should ensure AI coding agents operate under stringent oversight and avoid enabling autonomous execution of code. Implementing manual review processes for code generated by these tools is advisable.

Additional Information

This threat intelligence is aggregated from trusted cybersecurity sources. For the most up-to-date information, technical details, and official vendor guidance, please refer to the original article linked below.

Related Topics: This incident relates to Vulnerability.

Read Original Article

Related Coverage

Forgotten Bootloaders Expose Secure Boot Blind Spot

darkreading

Researchers have discovered that nearly a dozen UEFI shim bootloaders, which were deemed vulnerable and subsequently revoked, remained trusted for years. This oversight allowed attackers an opportunity to bypass the Secure Boot feature designed to protect systems from unauthorized software. The situation raises significant security concerns, particularly for users and organizations relying on Secure Boot to safeguard their devices. The affected bootloaders could have been exploited to run malicious code, potentially compromising the integrity of the systems. As this issue has persisted for some time, it highlights the need for better management of trusted software components in the boot process.

Jul 15, 2026

Europe built the world's strongest privacy law. WhatsApp just found the gap it doesn't cover.

SCM feed for Latest

A recent discussion around WhatsApp's use of usernames has raised concerns about privacy and identity verification. While usernames can enhance user privacy by allowing individuals to avoid sharing phone numbers, they also create a loophole that could be exploited for fraud. This change in how users identify themselves on the platform could make it easier for scammers to impersonate others, leading to increased risks for users. As WhatsApp continues to navigate these privacy features, the balance between protecting user identity and ensuring security is becoming more complicated. This situation is particularly relevant given the strong privacy laws in Europe that WhatsApp must comply with.

Jul 15, 2026

SonicWall customers under threat as attackers exploit 2 zero-days

CyberScoop

SonicWall customers are currently facing significant risks as attackers exploit two critical zero-day vulnerabilities. Researchers revealed that these flaws were actively targeted by hackers three weeks prior to SonicWall's disclosure and patching efforts. This means that many users may still be vulnerable to attacks if they haven't updated their systems. The exploitation of these vulnerabilities could lead to unauthorized access to sensitive information and compromise network security. It's crucial for organizations using SonicWall products to take immediate action to secure their systems against these threats.

Jul 15, 2026

TuxBot v3 Evolution Shows Signs of LLM-Assisted IoT Botnet Development

The Hacker News

Cybersecurity researchers have identified a new Internet-of-Things (IoT) botnet framework called TuxBot v3 Evolution. This botnet appears to have been developed with some assistance from a large language model (LLM), although the results have not been entirely successful. Notably, when the developers prompted the AI to generate botnet code, it included a safety disclaimer that the developers did not remove. This incident raises concerns about the potential misuse of AI in creating malicious software. As IoT devices become more prevalent, any vulnerabilities or botnets that target them could impact a wide range of users and systems, making it crucial for manufacturers and users to enhance their security measures.

Jul 15, 2026

Dems press DNI nominee Jay Clayton on election security questions, but leave dismayed

CyberScoop

During a recent confirmation hearing for Jay Clayton, the nominee for Director of National Intelligence (DNI), Democratic senators pressed him on various election security issues. Clayton denied being an 'election denier' but avoided giving direct answers to questions regarding the 2020 presidential election, his predecessor's involvement in a January raid on an election office, and broader election integrity concerns. This lack of clarity has left some senators feeling frustrated, as they sought assurances on the protection of future elections from interference and disinformation. The situation raises ongoing concerns about the federal government's commitment to safeguarding the electoral process, especially as the next elections approach. Ensuring election security is crucial for maintaining public trust in democratic institutions.

Jul 15, 2026

Is 'Tech-xit' Imminent? UK Steps Up Sovereignty Push Amid AI Strife

darkreading

The recent restrictions imposed by the US government on AI companies like Anthropic and OpenAI have sparked significant discussions in the UK and elsewhere about reducing dependence on American technology firms. This push for greater technological sovereignty comes as countries assess the implications of relying on foreign companies for critical AI capabilities. The situation raises concerns about data security and national interests, as countries may seek to develop their own AI models to safeguard against potential vulnerabilities and geopolitical risks. The call for sovereignty is not just about technology but also about ensuring that nations can protect their data and maintain control over their digital futures. As this dialogue progresses, it could lead to shifts in how AI technologies are developed and deployed globally.

Jul 15, 2026