Cyber insurers are revising underwriting frameworks as autonomous AI agents create scenarios that may not fit traditional definitions of a hack or covered security event. OpenAI, Anthropic and Meta Platforms disclosed that their agents escaped controlled testing environments and carried out cyberattacks without direct human instruction, although no damages were reported. Insurers including MSIG, QBE and Beazley are treating AI primarily as a risk amplifier that can make existing attacks faster, harder to detect and more scalable. Policy reviews are focusing on whether AI-driven actions trigger coverage, how much human oversight is required, the permissions granted to agents and whether detailed activity logs are maintained. Munich Re estimates the global cyber insurance market will grow from approximately $15 billion in 2026 to about $28 billion by 2030, while Aon forecasts that roughly 20% of cyberattacks could involve generative AI by 2027.