Marchner limits Caremark exposure—but makes effective AI oversight essential.
Executive Summary
Artificial Intelligence Governance – In Marchner v. B. Riley Financial, Inc., C.A. No. 2025-0164-LWW (Del. Ch. 2026), the Delaware Court of Chancery dismissed a stockholder derivative action that sought to characterize a failed investment and alleged warning signs at outside entities as a bad-faith oversight failure. The court reinforced a consequential limit on claims under In re Caremark International Inc. Derivative Litigation: directors’ oversight duties focus on legal compliance within the corporation, not on discovering misconduct at an unaffiliated counterparty before regulators do.
For boards overseeing artificial intelligence governance, the ruling provides a clear boundary—not a safe harbor. A vendor’s model failure, platform breakdown, or legal violation does not, standing alone, establish Caremark liability. The controlling question is whether the company has implemented and used an effective internal system to identify, assess, escalate, and report AI-related legal and compliance risk. Vendor diligence is a required component of that system; it is not a substitute for board-level oversight.
Key Takeaways for Directors and Senior Management
• Caremark is about internal oversight, not perfect foresight. The doctrine addresses bad-faith failures to establish or use board-level reporting systems for the corporation’s compliance with law. It does not make directors guarantors of a counterparty’s honesty, regulatory conduct, or business performance.
• Process remains decisive. The court credited allegations showing that B. Riley had an active Audit Committee, regular meetings, external auditors, and outside counsel. Gaps or imperfect results did not amount to an “utter failure” to implement a reporting system, and the plaintiff did not plead conscious disregard of internal legal violations.
• AI vendor risk must be governed from inside the enterprise. Management must inventory material AI uses, classify risk, conduct risk-based vendor diligence, secure appropriate contractual protections, monitor performance, and escalate material incidents. These functions must operate through a defined governance structure with accountable owners and a direct reporting line to the board or responsible committee.
• Documentation must establish operation, not merely design. Minutes, dashboards, risk-register updates, incident records, remediation decisions, and follow-up reports must demonstrate that the framework functions in practice. A charter, policy, or executive title provides little protection if material information does not reach the board and prompt a documented response.
The Decision
B. Riley invested substantial sums in connection with the 2023 take-private acquisition of Franchise Group, Inc. After the transaction closed, Franchise Group’s principal was implicated in an alleged securities-fraud scheme associated with Prophecy Asset Management LP, a separate third party. B. Riley later recorded significant write-downs. A stockholder filed a derivative complaint alleging, among other things, that directors ignored warning signs and faced a substantial likelihood of liability, making a pre-suit demand on the board futile.
The Court of Chancery rejected the effort to convert a disappointing investment into a non-exculpated loyalty claim. A Caremark plaintiff must plead particularized facts supporting an inference that directors acted in bad faith—typically by utterly failing to implement a board-level information system or by consciously ignoring red flags concerning unlawful conduct within the company. The alleged wrongdoing here occurred outside B. Riley. The court concluded that directors were not required to uncover third-party fraud before federal authorities and dismissed the action for failure to plead demand futility under Court of Chancery Rule 23.1. Marchner v. Riley, C.A. No. 2025-0164-LWW (Del. Ch. 2026).
What Marchner Means for AI Governance
The decision draws a legally significant distinction between an adverse AI outcome and a failure of oversight. A biased output, hallucinated result, cybersecurity event, service interruption, or vendor compliance failure does not by itself establish director bad faith. The decisive inquiry is whether the company operated a functioning process that identified the use case, assigned responsibility, assessed legal and operational risk, imposed controls, monitored the vendor, escalated material events, and enabled timely board action. That conclusion does not foreclose liability under contract, privacy, consumer-protection, employment, intellectual-property, cybersecurity, or sector-specific law; those exposures require separate analysis.
Recommended Actions
Boards and management teams should act now to: (1) integrate AI risk expressly into the enterprise risk management framework and assign accountable owners; (2) adopt a board or committee charter that fixes oversight responsibility and reporting cadence; (3) maintain a current inventory of material AI systems, including third-party and embedded tools, classified by risk; (4) require metrics-based reporting on compliance, testing, incidents, vendor performance, exceptions, and remediation; (5) establish escalation protocols that deliver material legal developments and incidents promptly to the responsible committee; and (6) align vendor diligence and contracting with the level of risk, including audit rights, incident-notification duties, data-use restrictions, security requirements, regulatory-cooperation obligations, liability allocation, and termination and transition rights.
Bottom Line. Marchner limits the circumstances in which third-party misconduct can support a Caremark claim. It does not relieve companies of the obligation to govern third-party AI risk through effective internal controls. The strongest defense is a documented, risk-based program that supplies directors with timely, decision-useful information and records the company’s response to material issues. Boards should test governance against the company’s actual AI deployment—not the language of its policies. This alert is for general informational purposes and does not constitute legal advice; application of the decision depends on the company’s facts, governing documents, industry, and regulatory obligations.
