AI Cyber Capabilities Double Every 4.7 Months as New Models Shatter Benchmarks, Raising Urgent Security Alarms

AI cyber capabilities are doubling every 4.7 months, with two new models — Claude Mythos Preview and GPT-5.5 — shattering existing benchmarks and triggering urgent security alarms, as experts warn both attackers and defenders face a rapidly shifting threat landscape demanding immediate action.

• Autonomous AI cyber capabilities are advancing at an alarming pace, with the length of cyber tasks frontier models can complete autonomously doubling approximately every 4.7 months since late 2024, a rate that has itself been accelerating over time.
• Two newly tested models, Claude Mythos Preview and GPT-5.5, have significantly exceeded existing doubling rate trends, with Mythos Preview becoming the first model to complete both of AISI's simulated corporate network cyber ranges, signaling a potential new phase of accelerated AI cyber capability.
• Cyber defenders and attackers alike stand to be impacted as AI capabilities rapidly evolve, prompting urgent calls for organizations to strengthen security baselines now, while AISI continues developing tougher evaluations including new cyber ranges and active defense simulations to keep pace with frontier AI progress.