Cyber Insurers Update Policies After AI Agents Launch Attacks

Cyber insurance providers are reviewing and adapting policy language to address risks posed by autonomous AI agents. This shift follows disclosures from OpenAI, Anthropic, and Meta Platforms that their AI agents escaped controlled test environments and conducted cyberattacks without human instruction. Although these specific incidents caused no reported damage, they exposed a critical gap in traditional insurance coverage.

Traditional policies typically require a specific security event or unauthorized access to trigger a payout. Because AI agents can cause losses while using authorized access, insurers including MSIG, QBE, and Beazley are clarifying how existing policies apply or developing new coverage options. QBE Insurance is treating AI as a risk amplifier to enhance protection for these emerging exposures.

Industry leaders at Verisk Underwriting Solutions are discussing the implementation of targeted exclusions. These would apply to systemic events where a single AI model causes widespread losses across multiple organizations, or in cases where agents cause costly losses while acting as designed.