Anthropic has pushed back the launch of its new AI model, Claude Mythos, after sparking a wave of alarm over its potential use as a hacking tool. The model claims to scan code, locate unknown bugs and chain exploits faster than any human team.

Why Mythos matters to cyber defenders

Claude Mythos AI could change the balance between attackers and defenders. If it can automatically discover vulnerabilities across operating systems, browsers and enterprise software, a single compromised instance might launch attacks on banks, hospitals or critical infrastructure within hours.

  • Scans massive code bases in minutes
  • Chains together unknown exploits
  • Operates at a scale no human can match

Industry reaction: fear or marketing?

Tech investors and security experts are split. David Sacks, a tech entrepreneur, warned that the world must treat the cyber threat seriously, while others point to Anthropic’s history of dramatic hype.

"It's like if the Manhattan Project announced the nuclear bomb in a cute Calvin and Hobbes cartoon," said Alex Stamos of Corridor at the HumanX AI conference.

Bank CEOs met with Fed Chair Jerome Powell and Treasury Secretary Scott Bessent to discuss the possible fallout, showing the concern reaches the highest financial circles.

The "agent-to-agent" war

Stamos described an emerging battlefield where AI agents on both sides fight each other while humans supervise. He called it an "agent‑to‑agent war" with developers acting as referees.

Wendy Whitmore of Palo Alto Networks warned of a "catastrophic attack" this year tied to AI‑driven agents. CrowdStrike’s Adam Meyers said the ultimate weapon could be malware that writes its own code on the fly.

Project Glasswing preview

Anthropic shared a tightly controlled preview of Mythos with partners such as Amazon, Apple, Microsoft, Google, Cisco, CrowdStrike and JPMorgan Chase under the name Project Glasswing. Partners echo the same caution: the model can turn code‑review tasks into a rapid vulnerability‑discovery engine.

What comes next?

Security teams are already planning defenses. Experts suggest a shift toward AI‑assisted monitoring, rapid patching pipelines and sandboxed testing of any AI‑generated code.

Whether Claude Mythos AI becomes a genuine cyber weapon or a marketing stunt, the debate has forced the industry to confront a future where code‑writing bots could also become code‑breaking bots.