Anthropic Claude Mythos sparks security fears as Powell, Bessent warn bank CEOs
Anthropic decides not to release Claude Mythos for public use
The US Treasury Secretary Scott Bessent and Federal Reserve Chair Jerom Powell have issued a dire warning to the financial sector, especially bank CEOs regarding AI-driven cyber risks.
According to the sources privy to matter, reported by Reuters, the Treasury convened the meeting in Washington, aiming to make key industries and the financial sector aware of the security risks posed by Claude Mythos and other models.
In the meeting, the CEOs were instructed to take effective steps against cyber threats to defend their systems.
According to sources, invitations for the event were sent while most major U.S. bank CEOs were already in Washington for other business. While leaders from Citigroup, Morgan Stanley, Bank of America, Wells Fargo, and Goldman Sachs attended, JPMorgan CEO Jamie Dimon was reportedly unable to join.
Recently, Anthropic has developed Claude Mythos Preview, a frontier model with unprecedented capabilities in finding and exploiting software and cybersecurity vulnerabilities.
According to the US-based artificial intelligence company, the latest model is highly capable of identifying and misusing vulnerabilities across “every major operating system and every major web browser.”
Given Claude Mythos’ unprecedented cyber capabilities, the model can overshadow skilled humans to find these weaknesses.
Earlier this week, Anthropic took a proactive step and briefed senior US government officials and key industry stakeholders on potential capabilities of the AI model.
Claude Mythos: ‘Pandora Box’
Anthropic has managed to develop such a powerful AI model, but it could cause AI-driven cyber threats. The model autonomously found a 27-year-old vulnerability in OpenBSD and a 16-year-old flaw in FFmpeg that had been overlooked in millions of automated tests.
When it comes to the Cybersecurity Vulnerability Reproduction benchmark, Claude Mythos (83.1 percent) outperformed Claude Opus 4.6 (66.6 percent).
However, the model also discovered and chained multiple vulnerabilities together to gain full system control without human intervention, leading to autonomous exploitation.
In the wake of unsettling and unregulated capabilities, Anthropic has decided to release the model for a handful of tech companies, such as Microsoft, Google, and 40 other organizations under the project named “Project Glasswing.”
Claude Mythos Preview will be accessible through the Claude API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry.
