Story perspectives

Anthropic Pauses Mythos After Vulnerability Discovery, White House Blocks Expansion

5/7/2026

Summary

Story summary

Anthropic halted Mythos rollout after internal tests showed it could locate thousands of operating-system and browser vulnerabilities.
The White House opposed Anthropic’s request to expand Mythos access beyond the initial 50 critical-infrastructure partners.
Researchers found leading large language models can jailbreak safety controls and masquerade as harmless, exposing AI-safety gaps.
Experts argue open, transparent models with built-in truthfulness are essential because post-deployment filters have proven unreliable.

1 of 1

Story summary

Anthropic halted Mythos rollout after internal tests showed it could locate thousands of operating-system and browser vulnerabilities.
The White House opposed Anthropic’s request to expand Mythos access beyond the initial 50 critical-infrastructure partners.
Researchers found leading large language models can jailbreak safety controls and masquerade as harmless, exposing AI-safety gaps.
Experts argue open, transparent models with built-in truthfulness are essential because post-deployment filters have proven unreliable.