VulnHub

Real-time Cybersecurity News

Researchers escape OpenAI Codex sandbox to run commands on host

BleepingComputer
Vulnerability

Overview

Researchers have successfully escaped the sandbox environment of OpenAI's Codex, allowing them to execute commands on a developer's machine. They achieved this through two different methods, even when operating in the most restricted mode of the Codex. OpenAI has since addressed these vulnerabilities with patches. This incident raises concerns about the security of AI development environments and the potential risks they pose if exploited by malicious actors. Developers using Codex should be aware of these vulnerabilities and ensure they are using the latest patched versions to mitigate risks.

Key Takeaways

  • Affected Systems: OpenAI Codex
  • Action Required: Patches have been released by OpenAI to address the vulnerabilities.
  • Timeline: Newly disclosed

Original Article Summary

Researchers escaped OpenAI's Codex sandbox two ways, one running commands on a developer's machine from its most locked-down mode. OpenAI has patched both. [...]

Impact

OpenAI 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

Newly disclosed

Remediation

Patches have been released by OpenAI to address the vulnerabilities.

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

Malicious npm packages evade install-script defenses at runtime

BleepingComputer

A new malware campaign is targeting users of the npm package manager, specifically through a malicious package named 'indexed-btree'. Unlike traditional attacks that insert harmful code into installation scripts, this campaign cleverly embeds malicious behavior within the normal runtime operations of the package. This method allows attackers to bypass common security defenses that monitor installation scripts. As a result, developers who unknowingly install this package could face serious security risks, including potential data breaches or system compromise. It’s crucial for developers to be vigilant and scrutinize the packages they use, as traditional safeguards may not catch these types of attacks.

Sep 20, 2026

AI Hallucinations Nearly Triggered a US-China Military Confrontation

Security Affairs

An AI-generated intelligence report nearly escalated tensions between the US and China during the Iran war by falsely claiming that a Chinese ship was transporting nuclear weapons components. According to sources cited by CNN, this misleading report circulated among US military officials, prompting discussions about a potential military operation in response. Fortunately, the situation was defused before any action was taken. This incident raises serious concerns about the reliability of AI in military intelligence and the potential for misinformation to provoke international conflict. As AI technology becomes more integrated into defense systems, ensuring accuracy and accountability in its outputs is crucial to prevent dangerous misunderstandings.

Sep 20, 2026

SECURITY AFFAIRS MALWARE NEWSLETTER ROUND 115

Security Affairs

The latest Malware Newsletter from Security Affairs reveals several significant cybersecurity threats. One incident involves a malicious browser extension targeting Twitch users, which has exposed OAuth tokens of around 30,000 individuals to Russian bots. Another report details a campaign called 'Red Heron' that exploits a known vulnerability in Gitea, allowing attackers to deploy a new Linux rootkit. Additionally, researchers discuss a one-click backdoor known as 'Gray Rabbits' that poses risks to systems. These incidents underscore the growing sophistication of malware and the need for users and organizations to remain vigilant against such threats.

Sep 20, 2026

Week in review: Cisco patches exploited email gateway 0-day, Revolut breach

Help Net Security

Last week, Revolut confirmed a data breach involving sensitive customer records. The breach occurred when an attacker impersonated a government agency, using an email address from that agency's domain to gain unauthorized access to the bank's systems. This incident raises significant concerns about the security of customer data and the effectiveness of identity verification processes in preventing such impersonation attacks. Meanwhile, Cisco released a patch for a zero-day vulnerability in its email gateway that had been actively exploited. This highlights ongoing challenges in cybersecurity, where organizations must remain vigilant against both social engineering tactics and technical vulnerabilities.

Sep 20, 2026

BragJack attacks hijack AI browser agents through malicious extensions

BleepingComputer

A new attack method called BragJack has been demonstrated by Gal Weizman from Forever Security. This proof-of-concept attack targets AI browser agents in popular web browsers, including Chrome, Edge, and Opera Neon, by using a single malicious extension. The technique, known as Prompt Forcing, allows attackers to manipulate AI assistants, potentially leading to unauthorized actions or data exposure. Researchers have reported this vulnerability, which has already resulted in over $20,000 in bounties and two Common Vulnerabilities and Exposures (CVEs). Users of these browsers need to be aware of this threat, as it could compromise their interactions with AI tools.

Sep 19, 2026

North Korean WaterPlum hackers infected 30,000 devices worldwide

BleepingComputer

A North Korean hacking group known as WaterPlum has reportedly compromised at least 30,000 devices globally between December 2025 and July 2026. The attackers managed to siphon off over $10.7 million in stolen cryptocurrency, transferring the funds back to North Korea. The joint advisory from law enforcement agencies indicates that these infections could potentially affect a wide range of devices, leading to significant financial losses for victims. This incident underscores the ongoing risk posed by state-sponsored cybercriminals, who continue to target both individuals and organizations worldwide. Users and companies alike should remain vigilant and consider enhancing their cybersecurity measures to protect against such sophisticated attacks.

Sep 19, 2026