Story perspectives

Anthropic's Mythos AI Shows 73% Hacking Success Rate

4/17/2026

Summary

Story summary

Anthropic's Mythos AI model is being tested under the Project Glasswing initiative with limited access to select organizations for cybersecurity purposes.
Mythos has demonstrated a 73% success rate in expert-level hacking tasks.
Cybersecurity experts remain skeptical about the threat's severity, saying its real-world impact may be overstated.
They warn vulnerabilities could be exploited more easily, but overall harm may be less than claimed.

History

Historical Context

In July 1996, President Bill Clinton signed Executive Order 13010 to combat rising cyber threats, following Senate hearings on computer security. This initiative established a task force involving agencies like the NSA, responding to a reported 50% increase in cyberattacks that year. Today, Anthropic's Mythos AI model under Project Glasswing reflects ongoing efforts to address similar cybersecurity challenges amid expert skepticism about technological vulnerabilities.

1 of 2

Story summary

Anthropic's Mythos AI model is being tested under the Project Glasswing initiative with limited access to select organizations for cybersecurity purposes.
Mythos has demonstrated a 73% success rate in expert-level hacking tasks.
Cybersecurity experts remain skeptical about the threat's severity, saying its real-world impact may be overstated.
They warn vulnerabilities could be exploited more easily, but overall harm may be less than claimed.

1 / 2