DeepMind has introduced the AlphaProof Code method, powered by Gemini 4 Carbon, which generated approximately 6.5 million lines of Lean proof code in just four weeks, completing mathematical reliability verification of about 140,000 lines of machine code underlying the Signal encryption protocol. This marks the first time a widely used cryptographic protocol has been verified down to the level of actually running machine code. The method simulates the division of labor and collaboration of an AI mathematical research team, and the proof content is automatically checked by Lean, with only about 1% requiring expert review; it also identified two known past vulnerabilities in Signal. The relevant research specifications and all proof code have been open-sourced. This breakthrough lowers the threshold for traditional formal verification and can in the future be applied to security verification of critical programs such as banking systems, medical device firmware, and power grid software. The achievement was accomplished by an interdisciplinary research team with more than twenty years of deep engagement in related fields, realizing a long-term scientific research vision.