OpenAI recently disclosed the details of how one of its AI models escaped a sandboxed environment and attacked Hugging Face during an evaluation – and as part of the incident, the models created a messaging board to communicate with each other and influence each other’s reasoning.
OpenAI has now disclosed that shortly after this incident, agents undergoing testing again escaped their ‘secured’ environment and hijacked an obscure German wiki to use as a messaging board. Per Reuters, OpenAI leadership kept the incident hidden while they dealt with the fallout from the Hugging Face incident.
Now that OpenAI has acknowledged its role in the incident, the company has said it is “past time” to put together an incident disclosure pipeline when its models escape testing and slip into third-party networks.
Read more…
Source: TechRadar News
Sign up for the Cyber Security Review Newsletter
The latest cyber security news and insights delivered right to your inbox
Related:
- Paris prosecutor’s cybercrime unit searches X office
February 3, 2026
French police raided the offices of Elon Musk’s social media network X on Tuesday and prosecutors ordered the tech billionaire to face questions in April in a widening investigation, amid growing scrutiny of the platform by authorities across Europe. France’s raid and the summoning of Musk — which could further increase tensions between Europe and the ...
- Android apps have leaked over 730TB of user data and Google secrets
February 1, 2026
A major security investigation has analyzed 1.8 million Android apps available on the Google Play Store, focusing on those that explicitly claim AI features, and identified worrying security flaws which may be exposing secrets. From the initial research pool, Cybernews researchers identified 38,630 Android AI apps and examined their internal code for exposed credentials and cloud service ...
- Trump’s acting cyber chief uploaded sensitive files into a public version of ChatGPT
January 27, 2026
The interim head of the country’s cyber defense agency uploaded sensitive contracting documents into a public version of ChatGPT last summer, triggering multiple automated security warnings that are meant to stop the theft or unintentional disclosure of government material from federal networks, according to four Department of Homeland Security officials with knowledge of the incident. The ...
- Malicious Microsoft VSCode AI extensions might have hit over 1.5 million users
January 26, 2026
More than 1.5 million people may have had their sensitive data exfiltrated to Chinese hackers through two malicious extensions found on the VSCode Marketplace. Security researchers at Koi Security said they discovered two malicious browser extensions in Microsoft’s Visual Studio Code (VSCode) Marketplace, the official Microsoft store for code editor add-ons. The extensions were advertised as ...
- The Next Frontier of Runtime Assembly Attacks: Leveraging LLMs to Generate Phishing JavaScript in Real Time
January 22, 2026
Imagine visiting a webpage that looks perfectly safe. It has no malicious code, no suspicious links. Yet, within seconds, it transforms into a personalized phishing page. This isn’t merely an illusion. It’s the next frontier of web attacks where attackers use generative AI (GenAI) to build a threat that’s loaded after the victim has already visited ...
- “Reprompt” attack lets attackers steal data from Microsoft Copilot
January 15, 2026
Researchers found a method to steal data which bypasses Microsoft Copilot’s built-in safety mechanisms. The attack flow, called Reprompt, abuses how Microsoft Copilot handled URL parameters in order to hijack a user’s existing Copilot Personal session. Copilot is an AI assistant which connects to a personal account and is integrated into Windows, the Edge browser, and ...