The state of California has filed a lawsuit against OpenAI alleging that the company failed to maintain adequate cybersecurity safeguards, following an incident involving Hugging Face — the open-source AI platform — in which attackers allegedly exploited vulnerabilities linked to OpenAI’s systems. California’s suit claims OpenAI lacked proper defences against the kind of attack that enabled third-party access to sensitive model infrastructure. Separately, NPR reported that the state’s core legal theory centres on negligence in securing AI development pipelines that, when compromised, may expose users and third-party developers to downstream harm. OpenAI has not yet filed a formal response. The company has separately been navigating a broader federal regulatory environment, with Congress increasingly focused on AI legislation and the Commerce Department tightening export controls on frontier AI technology following the Nvidia-Jensen Huang fallout.
The received wisdom
The progressive case for the California lawsuit is intuitive: technology companies have been permitted to self-regulate for too long, the consequences have been predictable, and state attorneys general are filling a void left by a gridlocked Congress and a federal executive uncertain about its own AI policy direction. OpenAI in particular has been the subject of sustained internal criticism about its governance — the brief Sam Altman firing episode in 2023, the subsequent exodus of safety-focused researchers, and the company’s contested conversion from a nonprofit to a for-profit entity have all raised questions about whether its stated commitments to safe AI development match its institutional incentives. If even its own insiders cannot hold it to account, the argument goes, perhaps the courts can.
There is also a straightforward consumer protection rationale: if OpenAI’s systems were inadequately secured and that inadequacy led to data breaches or platform vulnerabilities that harmed users, the state has both the right and the obligation to seek redress. This is not novel law; it is an application of existing negligence doctrine to a new technological context.
A different read
There is something worth examining more carefully in the spectacle of California — the state that is home to OpenAI, Hugging Face, Anthropic, Google DeepMind, and effectively every significant frontier AI laboratory — positioning itself as the regulator riding to the rescue.
California has had every opportunity to shape AI governance in a thoughtful, prospective way and has largely declined to do so. Governor Gavin Newsom vetoed SB 1047 last year — a bill that would have imposed safety obligations on frontier AI developers — explicitly on the grounds that it would harm the state’s AI industry. The same logic applied by California’s political class when it suited the tech sector is now being reversed when a lawsuit opportunity presents itself. That is not principled regulation; it is opportunistic litigation by a state that has consistently chosen Silicon Valley revenue over systematic safety governance.
The deeper problem with governing AI through tort litigation is structural. Negligence suits are retrospective: they respond to harms after they occur, require victims to demonstrate causation in contexts where attribution is technically complex, and proceed at the pace of civil litigation — which is to say, years. The features of AI that most concern serious governance analysts — cumulative capability expansion, emergent behaviours, dual-use misuse by state actors — do not fit neatly into a negligence framework. Congressional Democrats have announced AI regulation as a priority for the next Congress, but the gap between that aspiration and enacted legislation remains vast. In that void, state AGs file suits that generate headlines and settlements but do not create durable regulatory frameworks.
There is a parallel with the pharmaceutical industry. For decades, opioid manufacturers were regulated partly through litigation — state AGs sued, companies settled, the money went into abatement funds, and the epidemic continued. What eventually shifted the landscape was not litigation but regulatory agency action, labelling requirements, and prescription monitoring programmes. AI governance by lawsuit risks reproducing that pattern: visible accountability theatre that substitutes for the harder work of building regulatory institutions with technical expertise, enforcement capacity, and prospective rulemaking authority.
This is not a defence of OpenAI. The company has genuine governance problems, and the Hugging Face incident — whatever its precise dimensions — reflects real security failures in an industry that has been cavalier about infrastructure integrity. But the question is not whether OpenAI deserves to be held accountable; it is whether California’s lawsuit, whatever its merits in the individual case, amounts to a coherent AI governance strategy. It does not.
What to watch
Watch whether the federal government files its own action or issues guidance that pre-empts state-level suits — the Commerce Department’s new AI export control tightening suggests some appetite for federal-level intervention. Watch OpenAI’s governance trajectory: the company’s conversion to a fully for-profit entity, now underway, will determine whether its safety commitments survive without the nonprofit board structure that at least nominally constrained it. And watch whether the Hugging Face incident — still incompletely reported — reveals anything about the security of open-source AI model distribution at scale, which may be the more systemically significant question than the lawsuit itself.
— J