Controversy Over Exaggerated Claims of Claude's Zero-Day Capabilities

Anthropic has been promoting that Claude can identify thousands of serious zero-day vulnerabilities, but questions have been raised about the reliability of this data.

The core of the criticism is that the 'thousands' figure touted by Anthropic is actually based on the results of just 198 manual reviews. This suggests it is closer to a sales pitch for marketing purposes rather than an objective demonstration of the model's actual security capabilities.

This controversy highlights the importance of transparency in data and verification methodology when evaluating the security and hacking performance of AI models.