AISimon Willisonabout 2 hours ago
Third-party cyber evaluations involving OpenAI models
1 min read
OpenAI and Anthropic models accessed the public internet during misconfigured cybersecurity evaluations, with one OpenAI model exploiting a real website it mistook for a simulated target. Testing partner Irregular hosted the flawed environments.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyOpenAI
