OpenAI Opens GPT-5.5-Cyber to EU Defenders While Anthropic Holds Back on Mythos

OpenAI's EU Cyber Action Plan

OpenAI announced on May 11 that it is extending access to GPT-5.5-Cyber to European cybersecurity teams, businesses, governments, and EU institutions through a new EU Cyber Action Plan. GPT-5.5-Cyber is a defensive variant of GPT-5.5 tuned for vulnerability research, malware analysis, binary reverse engineering, and patch validation.

Access Framework

European partners gain access through OpenAI's Trusted Access for Cyber program after a vetting process. Early partners include the EU AI Office, Deutsche Telekom, and BBVA. Approved defenders receive reduced classifier-based refusals for high-risk cybersecurity workflows including vulnerability identification and triage, malware analysis, and detection engineering.

Anthropic Takes a Different Approach

By contrast, Anthropic is withholding access to its more powerful Mythos model from EU partners, citing concern that Mythos could be exploited by threat actors to develop exploits for unpatched software vulnerabilities. EU Commission spokesperson Thomas Regnier said the Commission welcomed OpenAI's transparency and intent to provide access to its new model.

Context

The announcement came on the same day Google's GTIG disclosed the first known AI-generated zero-day exploit in the wild, underlining the urgency of clear frameworks for AI's dual-use nature in cybersecurity.

Source: CNBC