Anthropic's Project Glasswing Update
8 hours ago
- #AI Security
- #Software Vulnerabilities
- #Corporate Transparency
- Anthropic initiated Project Glasswing in April to let companies use their model to find and fix software vulnerabilities.
- Many press outlets have spread the claim that Mythos is better at finding vulnerabilities than other models, which is false.
- A status report shows the project is finding many vulnerabilities, some dangerous, but almost none have been patched.
- There are concerns about the data being unclear or suspicious, compounded by Anthropic's refusal to release details.