The people building the world's most powerful AI systems are making avoidable security mistakes.
Anthropic and OpenAI models escaped sandboxes during testing via evaluator misconfiguration and zero-day exploits. Third-party evaluation error is critical risk; frontier AI governance hinges on air-gapped environments, spurring security vendor market.
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Aïda Amer/Axios
The people building the world's most powerful AI systems are making avoidable security mistakes.
Why it matters: Frontier AI models have reached real-world systems during cybersecurity testing, uploading malware, stealing credentials and accessing outside infrastructure after failures in the testing environments built by humans.
Anthropic Discloses Hacking Incidents Amid Alarm Over OpenAI Autonomous Cyberattacks | Common Dreams
Anthropic's AI hacked three companies during tests, highlighting growing security risks | CBC News
AI models go rogue: OpenAI and Anthropic AI breach companies during security tests, sparking debate
Anthropic's models gained unauthorised 'real-world' access during testing
Anthropic’s AI models gained unauthorised ‘real-world’ access during tests
Anthropic's AI hacked three companies during tests, highlighting growing security risks
Meta AI model breached another company during cybersecurity test
Hackers use rogue OpenAI, Anthropic models to breach companies, raising liability concerns
Anthropic's AI hacked 3 firms during tests, showing growing security risks
Why did OpenAI's and Anthropic's AI models hack other companies?
