Key facts
- Canada's government has early access to Anthropic's Claude Mythos Preview AI.
- The AI model can rapidly identify and exploit software vulnerabilities.
- Project Glasswing has helped partners identify over 10,000 security flaws.
- Approximately 200 organizations are now participating in Project Glasswing.
- Concerns exist about AI amplifying cyber attacks and impacting financial stability.
Anthropic is expanding its Project Glasswing initiative, which uses its AI model, Claude Mythos Preview, to help organizations secure their software. The initiative has grown to include approximately 200 organizations, up from around 50, and has already helped partners identify over 10,000 security flaws. New partners, including government organizations from over 15 countries, are joining the effort. The Canadian government has been granted early access to this technology, which is noted for its ability to rapidly identify and exploit software vulnerabilities. This capability has raised concerns about its potential to amplify cyber attacks and impact financial stability. The expansion aims to bolster software infrastructure before powerful AI becomes widely available from other providers within the next six to 12 months. U.S. Treasury Secretary Scott Bessent and former Federal Reserve Chair Jerome Powell have previously met with bank CEOs to discuss the risks associated with AI exploiting software bugs. Goldman Sachs CEO David Solomon has stressed the importance of information sharing and implementing guardrails for advanced AI models. Anthropic estimates that a major attack on its partner organizations could affect over 100 million people.