Researchers find OpenAI’s GPT-5.5 achieves similar performance on cybersecurity tasks as Anthropic’s Mythos Preview, casting doubt on the supposed…

GPT-5.5 and Mythos Preview have been pitted against each other in various cybersecurity evaluations, with results suggesting that OpenAI’s model is just as capable as its competitor. In a series of tests conducted by the UK’s AI Security Institute (AISI), GPT-5.5 achieved similar performance on key tasks such as rev…

AISI has been evaluating frontier AI models, including GPT-5.5 and Mythos Preview, through 95 different Capture the Flag challenges designed to test capabilities in cybersecurity. On Expert-level tasks, GPT-5.5 passed an average of 71.4%, slightly outperforming Mythos Preview’s 68.6%.

While some have raised concerns about the supposed threat posed by Mythos Preview, OpenAI CEO Sam Altman has criticized what he calls ‘fear-based marketing’ in promoting limited releases for certain AI models. In an interview, Altman suggested that such tactics may be used to sell expensive services or solutions.

In response to concerns around cybersecurity risk, Anthropic restricted the initial release of Mythos Preview to ‘critical industry partners.’ OpenAI has also taken steps to address these issues through its Trusted Access for Cyber pilot program, which allows security researchers and enterprises to verify their iden…

What matters

• GPT-5.5 passed 71.4% of Expert-level cybersecurity tasks, outperforming Mythos Preview by a margin of error
• GPT-5.5 solved a challenging reverse engineering task in under 11 minutes at a cost of $1.73 in API calls
• GPT-5.5 matches Mythos Preview’s performance on the Last Ones test range

Why it matters

GPT-5.5 matches Mythos Preview’s performance on the Last Ones test range

This GenAI News article was prepared in original wording using reporting and materials published by Ars Technica. Source reference: https://arstechnica.com/ai/2026/05/amid-mythos-hyped-cybersecurity-prowess-researchers-find-gpt-5-5-is-just-as-good/.

Drafted by the GenAI News review pipeline.