Project Glasswing and Claude Mythos
TG AI News·April 7, 2026 at 7:21 PM·
Trusted Source
New Claude has found vulnerabilities in all major operating systems and browsers, with the total number of vulnerabilities measured in the thousands. The model can autonomously write exploits that would take experts weeks to develop.
In this regard, Anthropic is launching Project Glasswing — allocating one hundred million dollars in credits for software audits for more than 40 organizations, with further tokens available for purchase at $25/$125 per million tokens. This preview will not be made publicly available, but the company plans to release models of the Mythos level at some point.
Recently, Anthropic also began implementing new safeguards in Claude Code — they are now more actively checking what the user is trying to do and stopping potential hacks. Additionally, the company plans to release a new version of Opus soon with additional safeguards. This is justified by the fact that Opus was also able to find a number of vulnerabilities, including in Linux.