GEICO Settles Pennsylvania Cancellation Probe, Agrees to Follow AI Guidance

Pennsylvania Attorney General Dave Sunday announced on May 22, 2026, that his office had reached an agreement with GEICO over what it called alleged unfair or confusing auto insurance cancellations. The release says GEICO agreed to act consistent with the Pennsylvania Insurance Department’s guidance on insurer use of artificial intelligence systems, and Sunday said the agreement “ensures that the company’s use of new technology is done within industry standards.” Both are commitments about how the company will operate from here.

The investigation began with a complaint from a new GEICO policyholder in West Philadelphia. The company flagged her policy for its standard 60-day review of new customers, requested additional documentation under threat of cancellation, and then canceled the policy when the consumer believed she had submitted the required documents. The final communication did not make clear that the submission was inadequate. The result was that the policyholder drove without coverage.

Under the settlement, GEICO agreed to follow Pennsylvania Insurance Department guidance on insurer use of AI systems and to add a week to the document-submission window for new policyholders selected for review. Two further terms go to proof of residency: one form of verification now suffices where the company had asked for two, and a driver’s license qualifies when the address matches the policy. The company also committed to training customer service representatives on the new requirements and on the need for clarity throughout the process. The agreement is not an admission of a legal violation.

The enforcement action matters for two reasons. First, it shows that state regulators and attorneys general are scrutinizing AI not only for discriminatory outcomes but also for procedural confusion. A cancellation tool that is technically neutral can still be unfair if the notices and timelines are not designed around the consumer’s experience. Second, it demonstrates that AI governance in insurance must include the consumer-facing workflow, not just the model itself. The same model with different notice language and a longer submission window produces a different regulatory outcome.

Anywhere an AI-assisted underwriting, cancellation, or renewal process can produce the same friction is worth reviewing on the same terms. In particular, any adverse action generated or triggered by an automated system should include a clear explanation of what the consumer must do, by when, and what happens if the deadline is missed. The GEICO case suggests that courts and regulators will read the whole communication chain, from first notice to final letter, and weigh it alongside the model’s accuracy.

The settlement also points to a practical remediation path. Adding a week to the document window, accepting one form of proof rather than two, and clarifying final denial communications are low-cost changes that can materially reduce the risk of a consumer being left uninsured. Carriers should audit their current new-business review timelines for similar chokepoints. If a policyholder has already paid a premium, the burden of proof should be heavier than for a standard application, and the communication of any inadequacy should be unmistakable. A policy that is canceled because a consumer did not understand a request is a foreseeable error, and regulators are now treating it as such.

This enforcement also arrives alongside broader AI bias litigation. In California, a federal judge let disparate-impact claims proceed against Workday in July 2024 over allegations that its algorithmic applicant screening rejected qualified candidates on protected traits, and in May 2025 granted preliminary certification of a nationwide ADEA collective. The common thread is that vendors and carriers are being held responsible for the real-world effects of AI systems they deploy, even when the systems are not making the final decision.

Across these matters neither the vendor’s disclaimer nor the carrier’s reliance on it has functioned as a defense. Responsibility has attached to whoever the consumer experienced as the decision-maker, which for an insurance decision is the carrier, whatever produced the output. That allocation now sits underneath AI governance expectations as an assumption rather than an open question.