Mythos Preview Clears Both AISI Cyber Ranges as AI Offensive Capability Doubling Rate Signals Accelerating Risk

Mythos Preview has become the first AI model to complete both cyber ranges designed by the UK's AI Security Institute (AISI), establishing a new benchmark for automated offensive cyber capabilities. The achievement places Mythos Preview ahead of GPT-5.5, which solved only one of the two ranges, underscoring a widening gap in cyberattack proficiency across leading AI systems.

AISI's cyber ranges function as controlled evaluation environments that measure how effectively AI models can execute multistep cyber operations. The ability to clear both ranges implies Mythos Preview possesses sufficient reasoning, planning, and execution capability to navigate complex attack sequences autonomously. The February 2026 internal estimate that the length of cyber tasks AI models can complete has doubled every 4.7 months since late 2024 provides critical context for the pace at which these capabilities are advancing.

The trajectory raises significant questions about the readiness of current defensive frameworks to contend with AI-augmented threat actors. As models demonstrate sustained progress on standardized offensive benchmarks, the gap between cutting-edge capability and widespread deployment narrows. Security institutions face growing pressure to integrate AI-native threat assessment into existing protocols, particularly as the operational ceiling for automated attacks continues to climb.