As companies increasingly deploy artificial intelligence in hiring, pricing, compliance, customer service, underwriting, and core operations, the next wave of fiduciary-duty litigation may ask a familiar question in a new setting: did the board make a good-faith effort to oversee a material enterprise risk? Delaware courts have developed that theory under Caremark, including in cybersecurity cases such as Construction Industry Laborers Pension Fund v. Bingle, where claims were dismissed despite a major cyberattack because the pleadings did not show director bad faith. Texas courts are not bound by Delaware law, but its oversight cases may still guide how plaintiffs frame AI‑related governance failures.

Texas law starts from a director-friendly place. The Texas business judgment rule generally protects officers and directors from liability for acts within the honest exercise of business judgment and discretion. In Sneed v. Webre, 465 S.W.3d 169 (Tex. 2015), the Texas Supreme Court described the rule as protecting corporate fiduciaries from liability for decisions made in the honest exercise of judgment, while holding that the rule did not create a standing barrier to a closely held corporation shareholder’s derivative claims.  Likewise, Ritchie v. Rupe, 443 S.W.3d 856, 881 n.43 (Tex. 2014), reflected the Court’s reluctance to second-guess board decisions merely because they harm a shareholder or later appear unwise.

But AI may test the boundary between protected judgment and actionable inattention. A board’s decision to adopt an AI tool will usually look like a business decision. Yet where a board effectively delegates or defers its own judgment to AI outputs, without understanding their limits or ensuring meaningful oversight, the analysis may shift. A board’s total failure to understand, monitor, or respond to known AI risks may look different, especially where the AI system affects legal compliance, consumer safety, discrimination risk, financial reporting, or cybersecurity. AI systems can be opaque “black boxes,” can replicate biased historical data, and can produce confident but wrong outputs, making board-level oversight more important.

The 2025 amendments to the Texas Business Organizations Code should make ordinary “AI went badly” claims difficult. New Section 21.419 (SB 29) applies to publicly traded Texas corporations and private corporations that opt in. It presumes directors and officers act in good faith, on an informed basis, in the corporation’s interests, and in obedience to law and governing documents. A claimant must rebut that presumption and prove a breach involving fraud, intentional misconduct, an ultra vires act, or a knowing violation of law, pleaded with particularity.

But it may not end the inquiry where directors consciously ignore red flags about unlawful AI use, misrepresent AI controls, approve mission‑critical systems without any reporting structure, or effectively abdicate their own judgment to AI outputs if that conduct rises to the level of one of the statute’s enumerated exceptions, such as intentional misconduct or a knowing violation of law.

Texas’s new Responsible Artificial Intelligence Governance Act (TRAIGA) does not create a private right of action, but may also serve to define the regulatory backdrop against which boards evaluate AI-related legal and compliance risks.

Practical Takeaways:

1. Fiduciary-duty claims premised on AI are unlikely to succeed based on outcomes alone. Texas law continues to protect business decisions, even bad ones.
2. The likely litigation focus will be process, not technical perfection. Boards do not need to understand every model, dataset, or algorithmic design choice, but they should have a framework for identifying material AI uses and monitoring legal, compliance, cybersecurity, discrimination, and operational risks.
3. The danger zone is conscious inattention to known AI risks. SB 29 should protect good-faith business judgment, but it may not protect directors who ignore red flags, misrepresent AI controls, approve mission-critical systems without reporting structures, or effectively abdicate their judgment to automated outputs.