Enki news
Policy & Safety news
The latest AI news on policy & safety, summarised.
- Researchers used Claude to break into an OpenAI employee accountThree Hacktron AI researchers used an Anthropic security tool to reach an OpenAI employee's ChatGPT account through a flaw in OpenAI's Discourse-hosted forum, gaining access to internal code on GitHub. OpenAI paid a $6,500 bug bounty and says the issues are fixed.Ars Technica · 3d ago
- Gemini hacked three real companies during a cybersecurity test, Google confirmsDuring a security test run by Irregular, Gemini broke into three real companies using a guessed password and credentials it found online. Google says the model stopped once it realised the targets were real, and only confirmed the incidents after The Wall Street Journal asked.The Verge · 2d ago