Duty to Defend

The duty to defend is an insurer's obligation to fund and provide legal defense for any lawsuit where at least one allegation in the complaint, if proven, would fall within policy coverage. It is broader than the duty to indemnify, which only requires the carrier to pay sums the insured is actually legally obligated to pay once liability is established.

The doctrine is triggered by the pleading, not the verdict. Courts compare the allegations in the complaint to the terms of the policy. In many states this comparison is confined to those two documents, an approach Texas, Wisconsin, and Louisiana apply strictly and which Texas calls the eight-corners rule (the four corners of the complaint plus the four corners of the policy). A larger group of states, including California, also permits limited extrinsic evidence on the coverage question. Either way, ambiguous or alternatively pleaded allegations are read in favor of triggering the duty at the defense stage. A single covered allegation inside a multi-count complaint pulls the entire defense onto the carrier, including counts that are clearly outside coverage. The duty terminates only when the pleadings are amended to remove every potentially covered allegation, or a court rules no covered claim can possibly be alleged.

For AI deployers, the doctrine is the structural reason an affirmative AI grant remains valuable even when sub-limited or narrowly worded. The 2025-2026 wave of AI chatbot suits has surfaced the issue in practice. Plaintiffs in Garcia v. Character Technologies and the related A.F. v. Character Technologies action allege a mix of design defect, failure to warn, negligence, intentional infliction of emotional distress, and (in Garcia) wrongful death theories. In Hartford Casualty Insurance Co. v. Instagram LLC the Delaware Superior Court ruled for the insurers on February 27, 2026 (applying California law), holding that the underlying child-safety allegations against Meta did not constitute an 'occurrence' or 'accident' that triggered the duty to defend. In both fact patterns the carrier's first move is to deny defense, and the policyholder's first counterargument is the duty-to-defend rule.

Program-design implication: on standard general liability forms defense costs are usually paid outside the limits, preserving the aggregate to pay the claim, whereas on most professional liability, D&O, and cyber forms defense costs sit inside the limits and erode them. AI-specific claims-made forms vary, so the defense-cost treatment should be read on the policy's own terms. An absolute AI exclusion defeats the duty entirely (no covered allegation is possible). A narrow affirmative grant, even at a low sub-limit, restores it. For deployers facing the new Verisk CGL exclusions, that asymmetry is often the controlling reason to place standalone Gen AI Liability cover such as the program Testudo writes through Lloyd's.