Microsoft Publishes AI Code of Conduct Barring Cyberattacks, Deception, and Oversight Evasion

Skynet Chance (-0.09%): An explicit, model-level constraint against deception, self-reinforcement, collusion, and shutdown evasion targets exactly the loss-of-control failure modes central to Skynet scenarios, modestly lowering risk if enforced. The effect is limited because the document is voluntary policy rather than a demonstrated technical guarantee of compliance.

Skynet Date (+1 days): Endorsement of deliberate pacing and embedded evaluators suggests a slight slowdown in deploying unconstrained frontier agents, pushing dangerous-capability scenarios marginally later. The deceleration is small since Microsoft continues aggressive capability development in parallel.

AGI Progress (0%): The code of conduct contains governance commitments rather than capability advances, so it barely moves AGI progress. Its main signal is Microsoft's internal expectation that superintelligent systems arrive within a decade.

AGI Date (+0 days): Safety red lines and evaluator processes add mild friction to frontier model release cycles but do not constrain underlying research or compute scaling. The timeline effect on AGI arrival is therefore near negligible.