Where your policy may fall short on AI.
Generative AI has moved faster than the insurance market can update its wordings. This review breaks down the four exposure gaps your underwriter is pricing - and the endorsements that close them.
Hallucination liability
An LLM tells your user something false and they act on it.
Air Canada, DoNotPay, and multiple legal-tech vendors have been held liable in 2023–24. Canadian tribunals rejected the 'the AI is a separate entity' defence - a precedent Ontario courts are already citing.
AI E&O endorsement extending professional liability to model output.
Training-data IP claims
Every downstream fine-tuner is a target for the NYT v. OpenAI class of suits.
Standard Tech E&O and Cyber policies exclude intellectual-property infringement arising from model training. Without an affirmative carve-back, a single claim can exhaust your tower.
Affirmative IP-infringement coverage carved specifically for AI use.
Model bias & discrimination
Automated decision-making in hiring, lending or healthcare is regulated.
AIDA, the EU AI Act, and 18 US states now codify algorithmic-discrimination liability. Ontario's implied-warranty and human-rights frameworks apply in parallel.
Algorithmic-liability extension + regulatory defence costs inside the tower.
Agentic-AI blast radius
Autonomous agents executing code and sending payments compound severity.
When your agent can move money or mutate production systems, a single prompt-injection or tool-misuse event drives loss faster than any breach response team can contain.
Limit sized to agent capability, not headcount; tool-scope sub-limits and circuit-breaker clauses.
Six questions. Find your gaps.
Answer honestly - most policies silently exclude the things you're about to check.
Does your product generate text, images, code or other content using a generative model?
Do you ship agentic workflows that take actions (run code, call APIs, send payments) without a human approving each step?
Do you call third-party LLM APIs (OpenAI, Anthropic, Google, others) as part of your product?
Do you fine-tune or train models on data you don't own, or that may contain copyrighted or licensed content?
Does your model make or inform decisions about people - hiring, lending, healthcare, insurance?
Do you store, log or retain user prompts or model outputs that may contain personal information?
- Hallucination liability
- Agentic blast radius
- Third-party LLM dependency
- Training-data IP claims
- Model bias & discrimination
- Privacy & data retention
- Hallucination liability
- Agentic blast radius
- Third-party LLM dependency
- Training-data IP claims
- Model bias & discrimination
- Privacy & data retention
Does your product generate text, images, code or other content using a generative model?
An LLM that tells users something false - and they act on it - can leave you liable for the outcome.
Check where your policy stands on AI.
Share your AI use-cases and a Truly Insurance broker will walk through the endorsements your model actually needs.
Talk to an AdvisorOr build your stack in the Tower Builder