5/14/26 New evaluations from the UK AI Safety Institute suggest that frontier AI models are rapidly advancing in offensive cyber capability, with an early checkpoint of GPT-5.5 that shows it now reaches a similar level of performance on cyber evaluation as Mythos.
AI-driven offensive capability is accelerating faster than many organizations are prepared for, but the article also indicates that models could significantly improve defensive operations, vulnerability management, and resilience.
https://www.aisi.gov.uk/blog/our-evaluation-of-openais-gpt-5-5-cyber-capabilities