XBOW: GPT-5.5 Matches Anthropic's Restricted 'Mythos' in Penetration Testing — Now Open to All

XBOW·high signal

Security firm XBOW tested GPT-5.5's hacking capabilities and found it matches Anthropic's closely guarded Mythos model. Vulnerability miss rate dropped to 10% (from 18% on Opus 4.6, 40% on GPT-5). Black box GPT-5.5 now outperforms GPT-5 with full source code access, effectively killing XBOW's whitebox benchmark. Visual acuity hit 97.5%. The key difference: OpenAI shipped these capabilities to everyone while Anthropic keeps Mythos restricted. 60 points on HN.