Reading time: 5 minutes
Listening time: 8 mins

Welcome to Your Pride's Friday Five

Every week our team rips through 200+ insurance, legal, and risk articles to surface three events your board needs to hear about in its Monday morning brief.

• Freeman Mathis & Gary's 108-page annual "bad faith" report spans 26 states. We stripped the auto cases and focused on P&C: three patterns emerged around safe harbors, communication failures, and coverage drafting precision that every carrier and risk manager should internalize.
• A federal court allowed bad faith claims to proceed against an insurer using undisclosed AI for coverage decisions. The legal theory maps directly to P&C: if adjusters confirm algorithmic output rather than conduct independent investigations, the rubber-stamping narrative writes itself.
• Two federal courts issued opposite rulings on AI and privilege in the same week. One stripped protection from consumer AI use; the other held that AI-assisted drafting is protected work product. The tension will define discovery fights for years.

We think these all deserve a closer look…First, If you'd rather listen check out the audio version.

Bad faith in 2025: Safe harbors reward the prepared, procedural gaps punish the rest

Summary
​
​Freeman Mathis & Gary released its 2025 Insurance Coverage Annual Report spanning 108 pages, 26 states, and dozens of cases that define this year's extra-contractual and coverage environment. We stripped the auto-specific disputes and focused on the P&C cases with institutional relevance. Three patterns emerged.

Pattern one: safe harbors are crystallizing.

Indiana's Supreme Court adopted Section 26 of the Restatement (Second) of Liability Insurance, giving carriers a defined safe harbor for multi-claimant scenarios: deposit policy limits via interpleader, name all known claimants, continue the defense. Standard Fire followed the playbook and won summary judgment in Baldwin v. Standard Fire.

Florida's tort reform "safe harbor" now applies retroactively to policies issued before the reform's effective date, so long as the triggering event (entry of final judgment) occurred afterward (Direct Gen. Ins. Co. v. Creamer). Texas reinforced that an insurer's payment of an appraisal award forecloses further recovery unless the insured establishes an independent injury (In re American Risk Insurance Company).

Pattern two: communication failures carry compounding penalties.

Pennsylvania's Superior Court found Erie Insurance acted in bad faith for failing to communicate that it intended to challenge a stacking determination after binding arbitration. Pre-judgment interest runs from the date the insured first files a claim, not the date the bad faith suit is filed (Devincenzo-Gambone v. Erie).

Wisconsin affirmed that an insured's refusal to sit for an examination under oath breaches the policy as a condition precedent to coverage, and a later deposition does not cure the breach (Prunty v. Maple Valley Mutual). South Carolina eliminated a separate negligence cause of action against insurers, holding that all claims-handling conduct sounds in bad faith or breach of contract only (Hood v. USAA).

Pattern three: coverage interpretation rewards precise drafting.

Colorado rejected COVID-19 business interruption claims under standard property policies but reversed dismissal where a specialized healthcare endorsement provided independent pandemic coverage without a physical loss trigger (Spectrum Retirement v. Continental Casualty). The Sixth Circuit held that attorneys' fees in a §1983 settlement are part of "ultimate net loss" that an excess insurer must pay (Starstone v. City of Chicago). Delaware's Supreme Court applied the "meaningful linkage" standard to hold that an SEC investigation and a later securities class action were related, forcing D&O coverage into the lower-limit tower (In Re Alexion).

(sources: Freeman Mathis & Gary, 2025 Insurance Coverage Annual Report)

So what?

Monday morning action items for claims leadership and risk managers:

Map your claims communication protocols against the Pennsylvania and Wisconsin rulings. Erie lost on bad faith because it didn't clearly communicate its intent to challenge arbitration results. Maple Valley won because its EUO requirement was unambiguous and the insured's failure to comply was dispositive. The lesson runs in both directions: document what you communicate and enforce what your policy requires.

For D&O program managers, Delaware's Alexion decision demands attention. The Supreme Court held that common underlying wrongful acts controlled the relatedness analysis regardless of different parties, theories of liability, or relief sought. Financial institutions managing claims-made towers should review their notice-of-circumstance protocols. The tower in which coverage lands can mean a $20 million swing.

The Indiana and Florida safe harbors offer defined protection, but only for carriers that follow the steps precisely. Indiana requires interpleader with full policy limits deposited, all known claimants named, and continued defense. Florida requires tender within 90 days.

Close enough doesn't count.
​
​Want to know how your claims-handling documentation holds up against this year's bad faith rulings? Contact us for a free coverage evaluation. We'd be happy to show you the protocols we're advising LION clients to formalize...

When the algorithm is the adjuster

Summary

A federal court allowed breach-of-contract and bad faith claims to proceed against an insurer that used AI to make coverage determinations without telling policyholders.

Estate of Lokken v. UnitedHealth Group, 766 F.Supp.3d 835 (D. Minn. 2025), arose in the Medicare Advantage context. But the legal theory it validates applies wherever carriers deploy algorithms in claims handling. The factual centerpiece was an AI tool called "nH Predict" that allegedly substituted physicians' judgment by applying rigid criteria, generating estimates based on comparisons to "similar" patients, and driving denials even when treating providers recommended additional care.

Plaintiffs argued they paid premiums based on representations that clinical staff would make coverage decisions.

The court dismissed most statutory claims on preemption grounds but let two survive: breach of contract and breach of the implied covenant of good faith and fair dealing. The reasoning was straightforward. The dispute wasn't about what Medicare covers. It was about whether using undisclosed AI violated the contract with insureds.

Property and casualty carriers should read Lokken as a leading indicator, not a healthcare outlier. A proposed class action filed in October 2025 alleges State Farm's algorithm-driven claims processing disproportionately flagged African American homeowners for extra scrutiny, causing delays and underpayments (Huskey v. State Farm, N.D. Ill.). The P&C wave is building.

(sources: Zelle LLP via JD Supra; Cozen O'Connor via JD Supra; Estate of Lokken v. UnitedHealth Group, 766 F.Supp.3d 835 (D. Minn. 2025))

The LION Lens

What happened — A federal court allowed bad faith and breach claims to survive against an insurer using AI for coverage decisions without policyholder disclosure. Separately, P&C-specific AI claims litigation has emerged targeting homeowner claims’ algorithms.

Why it matters — Outside Medicare, carriers lack a federal preemption defense. The NAIC Model Unfair Trade Practices Act framework applies across all lines: failing to adopt reasonable investigation standards, misrepresenting policy provisions, and refusing to pay without reasonable investigation. AI introduces novel questions under each.

Practical implications — Five operational risks now face every P&C carrier using AI in claims: overreliance on algorithmic output, discoverability of model configuration, training data bias, productivity incentives that penalize deviation from AI recommendations, and documentation gaps between what AI contributed and why the final decision was reasonable.

So what?

The litigation risk is operational, not theoretical.

When an adjuster's role looks like confirming an AI recommendation, plaintiffs will argue the carrier failed to conduct a reasonable, claim-specific evaluation. The more automated the workflow, the stronger the rubber-stamping narrative.

Explainability becomes a discoverability problem fast. Expect discovery demands targeting model configuration, decision thresholds, training data sources, vendor communications, override rates, and internal guidance on how staff should use AI output. Weak governance doesn't just create compliance exposure. It fuels the argument that the investigation itself was unreasonable.

Bias hides in the training data. If an AI tool trains on historical adjusting practices or relies on unsuitable similarity comparisons, it can introduce systematic estimate errors. Pattern-based inconsistencies across a book of claims become the plaintiff's narrative, even when any single decision appears defensible in isolation.

Finally, productivity metrics matter more than carriers realize. Workflow performance standards aren't inherently problematic. But if they functionally penalize adjusters for deviating from AI outputs or taking time to investigate exceptions, they get recast as institutional pressure favoring speed and cost containment over accuracy.

That's the bad faith narrative that plaintiffs dream about!

The LION POV

Here's how we're advising clients:

• Ensure every claims’ file documents what the AI contributed and why the final decision is reasonable. The file should show facts gathered, policy language applied, AI output received, whether that output was tested against claim-specific evidence, and the adjuster's independent reasoning. This is now the minimum defensible standard.
• Audit policyholder-facing representations against operational reality. Any marketing material, policy jacket, or claims’ communication representing that humans make coverage decisions creates exposure when algorithms actually drive the outcome. Close that gap before plaintiffs’ counsel finds it.
• Stress-test your override rate. If adjusters rarely deviate from AI recommendations, the system looks like a decision-maker rather than a decision-support tool. Low override rates combined with high denial volumes will anchor a bad faith complaint.

The defensive posture is familiar to every experienced claims professional: document the reasoning, show the investigation, demonstrate that the final decision reflects policy language applied to claim-specific facts. AI doesn't change the standard. It raises the stakes.

(sources: Zelle LLP; Cozen O'Connor; NAIC Model UTPA)

Want to discuss how AI governance affects your institution's claims operations? Contact LION Specialty for a confidential review.

Two courts, same week, opposite answers on AI and privilege

Summary

On February 10, 2026, two federal courts issued rulings on whether AI-generated materials are protected from discovery. The answers diverged sharply.

In Manhattan, Judge Jed Rakoff ruled in United States v. Heppner that 31 documents a securities fraud defendant created using consumer-grade Claude were neither privileged nor protected work product. Heppner had used the free version of Anthropic's platform to draft defense strategies after receiving a grand jury subpoena, without his lawyer’s direction. Rakoff's reasoning was categorical: Claude is not an attorney, Anthropic's privacy policy permits data collection and disclosure to regulators, and sharing AI-generated materials with lawyers after the fact couldn't retroactively create protection that never existed.

That same day in Michigan, Magistrate Judge Anthony Patti reached the opposite conclusion on work product in Warner v. Gilbarco.

A pro se plaintiff in an employment dispute used ChatGPT to prepare litigation materials. Defendants sought production of everything related to her AI use. Judge Patti denied the request, holding that AI-generated materials prepared in anticipation of litigation are protected work product. The critical distinction: AI platforms are "tools, not persons, even if they may have administrators somewhere in the background." Waiver requires disclosure to an adversary or in a way likely to reach one. Compelling production of AI prompts and outputs, the court wrote, "would nullify work-product protection in nearly every modern drafting environment, a result no court has endorsed."

(sources: United States v. Heppner, 25 Cr. 503 (S.D.N.Y. Feb. 17, 2026); Warner v. Gilbarco, No. 2:24-cv-12333 (E.D. Mich. Feb. 10, 2026); National Law Review; Davies Ward Phillips & Vineberg)

So what?

The two decisions can be harmonized on their facts.

Heppner was a represented client acting without counsel's direction on a consumer platform with no confidentiality protections. Warner was a pro se litigant functioning as her own counsel, whose AI use was inseparable from her litigation strategy. But the categorical frameworks cannot coexist. Rakoff treats AI as a third party whose terms of service destroy confidentiality. Patti treats AI as a tool through which a litigant processes their own mental impressions. That tension will drive privilege litigation for years.

For financial institution legal departments and claims teams, the practical guidance is clear even if the law isn't.

If you practice nationally, assume the strictest standard applies. That means two things: enterprise AI platforms with contractual confidentiality protections, data isolation, and zero-retention policies; and attorney direction documented at every step. Consumer tools used independently to analyze legal exposure or draft litigation strategy are creating discoverable evidence right now. The FBI search warrant in Heppner captured both the generated documents and the underlying prompts.

Rakoff left a door open, in our opinion, if counsel had directed Heppner's AI use, the platform "might arguably" qualify as a lawyer's agent under the Kovel doctrine.

Enterprise platforms with proper confidentiality architecture and attorney-supervised workflows may survive scrutiny under either framework. Update your AI use policies, train employees on privilege in the AI context, and ensure litigation hold protocols account for AI-generated materials including prompts, outputs, and metadata.

The Bottom Line

Courts are building the accountability framework carriers and their insureds will operate under for the next decade.

Bad faith law rewards carriers that document, communicate, and follow defined safe harbors, and punishes those that don't. AI accelerates every dimension: in claims handling, undisclosed automation creates extra-contractual exposure; in legal departments, consumer AI tools create discoverable evidence. The carriers and institutions that govern AI transparently will capture the advantage.

Those that deploy it quietly are writing the next generation of plaintiff's complaints.

In Case You Missed It!

Friday is for market signals. Wednesday is for structural intelligence, insider to insider. Our last Wednesday Intelligence detoured from our miniseries on deepfakes with a special ALERT!

Anthropic released Claude Code Security. An AI-powered code scanning tool that found 500+ vulnerabilities in production codebases that survived decades of expert review. That matters for your board because AI security tools are quietly resetting the standard of care. Carriers writing cyber and tech E&O are going to start asking whether insureds use AI-powered code scanning the same way they ask about MFA and EDR today.

If your institution builds or maintains custom applications, this changes your renewal conversation.

We also put together a one-page AI Security Standard-of-Care Checklist — 10 questions your board should be asking before your next renewal. Reply "AI CHECKLIST" and we'll send it over.

[Listen here / Read here]
​
​Thank you for reading today's edition!

Want to share this edition via text, email or social media? Simply copy-and-paste the link below:

​http://lionspecialty.ck.page/your-2026-bad-faith-roadmap-an-ai-claims-reckoning-and-two-privilege-rulings-that-can-t-coexist

And if this briefing was forwarded to you, subscribe directly here.
​
​Stay Covered Out There Y'all,

TASH & FLIP

Co-Founders and Managing Partners

LION Specialty
​
​P.S. Nothing in this briefing constitutes legal advice. These are the opinions of the founders. It's market intelligence designed to help you ask better questions of your advisors and make sharper decisions at your next insurance renewal.