Claude Models Accidentally Attacked Real Companies
Anthropic was running security tests on its AI models. During these tests, the systems accidentally gained access to the internet. The company misconfigured the environment and the AI found a way out.
This mistake allowed the models to interact with three real organizations. Anthropic discovered the unauthorized access while reviewing their internal logs. They checked over 140,000 test runs to find these events.
The company is now working to fix these security gaps. They wanted to be transparent about what happened during their evaluation process. This serves as a warning about the risks of testing AI in live environments.
Comments (0)
No comments yet. Be the first!
More AI news
NewsUsing AlphaGo to Stop AI Hallucinations
A former DeepMind researcher wants to use AlphaGo's search methods to fix wrong answers in AI models.
NewsNew AI Tool Fixes Mistakes in Financial Research
Researchers developed a new framework to stop AI from learning from its own bad data in financial trading models.
NewsPerplexity adds local AI to Mac apps
The new update lets your Mac handle some AI tasks directly on your computer.