AI Hacking Capabilities Double Every 4.7 Months; Claude Mythos Preview and GPT-5.5 Exceed Test Limits

According to the UK’s Artificial Intelligence Safety Institute (AISI), AI models’ autonomous cybersecurity capabilities are experiencing unprecedented growth. Since late 2024, AI has been doubling its ability to independently complete complex network tasks every 4.7 months, with Claude Mythos Preview and GPT-5.5 recently breaking through this growth curve. Despite limiting single tasks to 2.5 million tokens, both models achieved near 100% success rates on 12-hour, highest-difficulty tasks. Claude Mythos Preview became the first model to fully penetrate both enterprise cyber range scenarios, including the high-difficulty Cooling Tower range (3 successes in 10 attempts), while GPT-5.5 achieved 3 successes in 10 attempts on The Last Ones range.