What to know about the AI models that are jolting Washington - Politico

Researchers who have tested Anthropic's Mythos and OpenAI's GPT-5.5 say their hacking capabilities are a 'game-changer' and advancing much faster than anticipated, according to interviews with nine top cyber researchers published by Politico on May 24, 2026. The models demonstrate advanced offensive cyber capabilities that could automate sophisticated hacking operations.

The article notes that these capabilities have 'jolted Washington,' raising concerns among policymakers about the implications for national security and cybersecurity defenses. The researchers emphasized the speed of advancement, stating that the models' hacking abilities are progressing more rapidly than expected.

The context comes amid ongoing debates about AI safety and regulation. The revelations add urgency to calls for new cybersecurity measures and oversight of advanced AI models with dual-use capabilities. Neither Anthropic nor OpenAI provided detailed public comment on the specific cyber capabilities described by the researchers.