Overview
- Anthropic kept its Claude Mythos Preview out of public release, giving access only to roughly 40–50 vetted partners to help find and fix software bugs the model can autonomously identify and combine into working exploits.
- After the announcement, the U.S. Treasury held urgent talks with major banks to gauge how fast-improving AI could expose holes in systems that move money, run transactions, and keep core services online.
- News reports say Vice President JD Vance and Treasury Secretary Scott Bessent joined a private call with CEOs from Anthropic, OpenAI, Google, Microsoft, xAI, CrowdStrike, and Palo Alto Networks to discuss AI security, a call described by people familiar with the matter.
- Security researchers warn that tools like Mythos could let more attackers find flaws faster, with an Anthropic lead saying similar capabilities could spread within six to 12 months, while other commentators question whether some claims overstate real-world exploitability.
- At the HumanX conference, investors and founders voiced strong demand for Anthropic’s enterprise products, even as the company navigates a Pentagon supply‑chain dispute in court and coordinates a partner program focused on rapid vulnerability remediation.