Particle.news

Dimon Warns Anthropic’s Mythos Could Pose Systemic Security Risk

The model’s skill at finding software flaws has prompted regulators to treat it as a potential systemic security threat.

Overview

  • JPMorgan CEO Jamie Dimon said at a July 15 Pennsylvania summit that broad access to Anthropic’s Mythos is dangerous, likening it to giving ballistic missiles to individuals and saying U.S. officials are now focused on the issue.
  • Mythos is an unreleased Claude-family model that scored exceptionally on April cybersecurity benchmarks by locating software vulnerabilities much faster than prior tools.
  • Anthropic has refused a public release and instead runs Project Glasswing, an invitation-only program that expanded from roughly 50 to about 150 partner organizations to use Mythos for defensive testing.
  • The company has tied usage to remediation by issuing $100 million in credits for vulnerability detection and patching, and major firms including JPMorgan and some crypto companies have sought access for security work.
  • Reported large funding and IPO planning have raised the stakes for governance and oversight because wider availability could increase the risk of AI-assisted exploitation, so markets and regulators are likely to scrutinize future access and controls.