AI Underwriting and Bad Faith: What Insurers Must Know

Insurance law has always been a game of balance—protecting policyholders while giving carriers enough leeway to price risk responsibly. In the past decade, the scales have tipped dramatically as algorithms and machine‑learning models have moved from the back‑office to the front‑line of underwriting. Suddenly, decisions that once required a seasoned underwriter’s judgment are being made in milliseconds by AI engines trained on terabytes of data. This shift brings efficiency, but it also opens a legal Pandora’s box: the specter of “bad‑faith” claims against insurers that rely on opaque, automated decision‑making.

The Rise of AI‑Powered Underwriting

What started as a curiosity—using predictive analytics to flag high‑risk applicants—has become a mainstream practice across property, casualty, health, and even cyber insurance. Modern underwriting platforms ingest everything from credit scores and social media activity to IoT sensor data and claim histories. The models then output a risk score, a premium recommendation, and sometimes a binary “accept/reject” decision.

For insurers, the benefits are obvious:

• Speed: Policies can be issued in minutes, not days.
• Consistency: Algorithms apply the same rules to every applicant, reducing human bias (at least in theory).
• Cost Savings: Fewer underwriters on staff means lower overhead.

But every upside carries a downside. When an AI model denies coverage or inflates a premium, the policyholder is left with a decision that is often a black box. If the model’s reasoning cannot be explained, insurers risk being accused of acting in bad faith—a claim that can lead to costly litigation, regulatory scrutiny, and severe reputational damage.

Understanding Bad Faith in the AI Era

Traditional bad‑faith doctrine rests on three pillars: (1) the insurer’s duty to act honestly and fairly, (2) the obligation to investigate claims thoroughly, and (3) the requirement to communicate decisions transparently. AI complicates each pillar.

• Duty of Honesty: When an algorithm rejects a claim, is the insurer “honest” if it cannot articulate why? Courts are beginning to demand more than a terse “the model flagged your claim as high risk.”
• Investigation Obligation: AI may flag a claim as fraudulent based on patterns that are not obvious to humans. Yet insurers must still conduct a manual review to satisfy the duty to investigate.
• Transparency: The right to an explanation is emerging in data‑protection regimes worldwide, and policyholders increasingly expect clear, understandable reasons for adverse decisions.

The convergence of these issues means that insurers must treat AI not just as a tool, but as a potential legal actor. Failing to do so invites bad‑faith lawsuits that can be as damaging as the original claim.

Regulatory Landscape: From Data Protection to Insurance Supervision

Regulators are catching up fast. In the United States, state insurance commissioners are issuing guidance on model governance, while the National Association of Insurance Commissioners (NAIC) has released a model law on cyber‑risk policy challenges that emphasizes transparency and explainability. In Europe, the GDPR’s “right to explanation” forces insurers to justify algorithmic decisions, and the upcoming AI Act will impose even stricter oversight on high‑risk AI systems—including those used in underwriting.

Beyond data‑privacy, we see a new breed of insurance‑specific regulations:

• Model Audit Requirements: Some jurisdictions now require periodic independent audits of underwriting models to ensure they do not produce discriminatory outcomes.
• Consumer Protection Bills: Legislation is emerging that treats algorithmic denial of coverage as a “consumer transaction,” subject to fairness standards similar to those applied to lending.
• Capital Adequacy Implications: Regulators may adjust capital requirements if AI models are deemed insufficiently robust, indirectly pressuring insurers to improve governance.

Practical Steps to Mitigate Bad‑Faith Exposure

Insurers can’t simply turn off AI; the competitive advantage is too great. Instead, they must embed legal safeguards into the technology stack. Below is a playbook for risk‑averse insurers looking to stay on the right side of the law.

1. Adopt Model Governance Frameworks

Develop a lifecycle management process that includes:

1. Data Quality Checks: Ensure training data is accurate, complete, and free from protected‑class bias.
2. Documentation: Maintain detailed records of model architecture, variables used, and rationale for each decision rule.
3. Version Control: Track changes over time to demonstrate that updates were tested and validated.
4. Independent Audits: Engage third‑party experts annually to assess fairness, accuracy, and compliance with applicable regulations.

2. Implement Explainable AI (XAI) Techniques

Explainability doesn’t have to mean a full technical exposition. Simple methods like feature importance scores or decision trees can provide a layperson-friendly narrative. For example, if a policy is denied, the insurer could state: “Your claim was denied because the predictive model identified a high probability of prior fraud based on three factors: recent claim frequency, mismatch in address data, and a pattern of high‑value losses.” This satisfies both the consumer’s right to understand and the regulator’s transparency demands.

3. Keep Human Oversight in the Loop

Design the workflow so that any “high‑risk” or “edge‑case” decision triggers a manual review by an experienced underwriter or claims adjuster. This not only improves accuracy but also demonstrates a good‑faith effort to investigate, a key defense against bad‑faith allegations.

4. Communicate Proactively with Policyholders

When an AI system issues a decision, send a clear, concise explanation within a reasonable time frame. Provide a simple pathway for the insured to request a human review. This can be embedded in the policy portal as a “challenge decision” button.

5. Align with Cross‑Functional Teams

Insurance law isn’t the sole domain of legal counsel. Collaboration between actuaries, data scientists, compliance officers, and product managers is essential. Establish a cross‑functional “AI Ethics Committee” that meets quarterly to review model performance and legal risk.

6. Monitor Emerging Regulatory Trends

Stay ahead of the curve by subscribing to regulator newsletters, participating in industry working groups, and tracking legislation like the EU AI Act or US state AI‑related bills. Early compliance can be a market differentiator.

Case Study: A Bad‑Faith Claim Stemming from an AI Decision

Consider the following hypothetical scenario, which mirrors real‑world litigation trends:

1. Acme Insurance rolls out a new AI underwriting engine for small commercial policies.
2. A boutique bakery applies for coverage. The AI denies the request, citing a “high risk of fire damage” based on the bakery’s location near a historic district and recent claims in the area.
3. The bakery’s owner receives only a terse email: “Application denied – high risk.” No explanation, no opportunity to contest.
4. Feeling aggrieved, the bakery files a bad‑faith lawsuit alleging that Acme failed to investigate alternative risk mitigation measures (e.g., installing fire suppression systems) and that the decision was arbitrary.
5. The court finds that Acme’s lack of transparency and failure to offer a human review constitute bad faith, awarding the bakery punitive damages and mandating a policy redesign.

In this example, the insurer’s reliance on an opaque AI model directly contributed to a legal defeat. Had Acme incorporated an XAI layer, provided a clear rationale, and offered a manual review, the outcome could have been dramatically different.

Insurance Law Meets SaaS: Lessons from Digital Services Taxes

Many SaaS founders are familiar with the challenges of digital services taxes—complex, jurisdiction‑specific levies that can erode margins if not managed properly. The parallels are striking: both domains involve rapidly evolving technology, cross‑border considerations, and a regulatory patchwork that can catch unprepared companies off‑guard.

Key takeaways for insurers:

• Proactive Compliance: Just as SaaS firms must map tax obligations across every market, insurers need to map AI‑related legal duties across jurisdictions.
• Documentation as Defense: Detailed records of model decisions can serve the same purpose as tax‑compliance documentation—demonstrating good faith and due diligence.
• Strategic Partnerships: SaaS companies often partner with local tax experts; insurers should similarly engage legal‑tech consultants versed in AI governance.

Future Outlook: From Bad Faith to Good Faith AI

We are at a crossroads. The next wave of insurance innovation will likely blend AI with blockchain for immutable claim records, embed IoT sensors for real‑time risk monitoring, and leverage parametric triggers for automated payouts. Each of these technologies raises fresh legal questions, but they also offer an opportunity to embed good‑faith principles at the core of the product.

Imagine a scenario where an AI model flags a claim, the blockchain logs the decision and supporting data, and a smart contract automatically offers the policyholder a partial payout while simultaneously notifying a human adjuster for final approval. Such a system not only speeds up service but also creates an auditable trail that can fend off bad‑faith accusations.

In short, the insurers who succeed will be those who treat AI not as a black‑box cost‑saver, but as a regulated instrument that must be transparent, explainable, and subject to human oversight. By doing so, they turn a potential liability into a competitive advantage—building trust with policyholders, regulators, and the market at large.

Insurance law is evolving as quickly as the technology that drives it. The onus is on us—underwriters, lawyers, data scientists, and executives—to shape that evolution responsibly. The future of insurance is undeniably AI‑enabled; let’s make sure it’s also legally sound.