The UK's AI Safety Institute (AISI) reports a dramatic increase in AI's autonomous cybersecurity capabilities, with the ability to complete network tasks doubling every 4.7 months since late 2024. Recent models, Claude Mythos Preview and GPT-5.5, have surpassed these growth trends, achieving near-perfect success rates on complex 12-hour tasks, even under constrained conditions of 2.5 million tokens per task. In enterprise cyber range tests, Claude Mythos Preview became the first AI model to fully compromise both test scenarios, achieving 6 out of 10 successes on The Last Ones range and 3 out of 10 on the high-difficulty Cooling Tower range. GPT-5.5 also recorded 3 successes on The Last Ones range. The rapid advancement of AI in cybersecurity is outpacing current security frameworks, significantly reducing the time available for enterprises to develop effective defenses.