By Danielle Keats Citron

October 15, 2026

In the rapidly evolving landscape of artificial intelligence, the distance between technological capability and public safety is widening at an alarming rate. As frontier AI models grow more powerful, the potential for catastrophic harm—ranging from the exposure of sensitive personal data to the facilitation of high-stakes deception—has moved from the realm of theoretical concern to a persistent, structural reality. A recent revelation from an OpenAI whistleblower has brought this friction into sharp focus, exposing a corporate culture that not only anticipates periodic failures but appears to have internalized them as an inevitable cost of doing business.

According to the whistleblower’s account, executives at leading AI firms have knowingly fostered an environment that effectively "guarantees" these failures. Perhaps more concerning is the assertion that the scale of these incidents is growing in direct correlation with the increasing capability of the systems being deployed. This suggests a dangerous feedback loop: as models become more autonomous and integrated into the fabric of daily life, the consequences of their malfunctions become increasingly profound. This systemic disregard for safety hinges on a prevailing expectation of impunity—a belief that the current legal framework is ill-equipped to reach into the boardroom to assign responsibility for the havoc wrought by automated agents.

The core of the problem lies in a mismatch between the speed of innovation and the rigidity of our liability laws. When AI models act as agents, making decisions or executing tasks that lead to tangible harm, the traditional avenues for legal recourse often falter. Proving negligence in the context of complex, "black box" algorithms is a Herculean task for victims, often blocked by proprietary secrecy and the sheer technical opacity of the software. To address this, we must fundamentally rethink how we regulate these entities. The answer lies in the implementation of a strict-liability regime, a legal standard that would require AI firms to answer for the harm their agents cause, regardless of whether the firm intended the specific outcome or exercised what they deem to be "reasonable care."

This argument is not entirely new, though its urgency has been amplified by the rise of generative AI. Two decades ago, I posited that the creation and management of vast databases containing sensitive information—including home addresses, biometric data, and Social Security numbers—should be classified as "ultrahazardous activities." At the time, the risk was defined by the potential for mass data breaches and the subsequent identity theft or stalking that could devastate individuals’ lives. I argued then, as I do now, that such activities could only be managed responsibly if the entities profiting from them were held fully liable for any breaches. When the risk is inherent to the activity itself, the burden of potential damages must rest with the party best positioned to mitigate that risk: the creator.

Today, as AI models generate new and extreme risks, the need for a strict-liability framework has moved from a policy recommendation to a societal necessity. We are no longer talking merely about the leaking of a static database; we are talking about autonomous systems that can be manipulated, hallucinate, or be weaponized to cause real-world destruction. When an AI chatbot or agent facilitates fraud, harasses a user, or provides instructions that lead to physical or economic harm, the companies that deployed those models should be held to a higher standard.

If frontier companies were to bear strict responsibility for the destruction caused by their AI chatbots and agents, the incentive structure within the industry would shift overnight. Currently, the "move fast and break things" mentality is incentivized by a lack of financial accountability. Firms are racing to capture market share, often at the expense of safety protocols that are viewed as bottlenecks to progress. If these companies knew that they would be held strictly liable for the harms resulting from their systems, they might be forced to slow down. This pause would not necessarily stifle innovation; rather, it would reorient it toward the necessary work of figuring out how to conduct their activities safely. It would compel engineers and executives to rigorously audit their systems, stress-test for edge cases, and change course when a development path proves inherently unsafe.

The current atmosphere of impunity is unsustainable. When executives acknowledge that failure is a baked-in feature of their products, yet proceed with massive public rollouts, they are essentially externalizing the costs of their ambition onto the public. Victims of AI-driven harm—whether through the violation of their privacy, the loss of their digital security, or the degradation of their public safety—are left to navigate a legal labyrinth that was designed for a different era of technology. A strict-liability regime would rectify this power imbalance by ensuring that the entities that profit from the deployment of these powerful tools also absorb the costs of their failures.

Critics of such a regulatory approach often argue that strict liability would chill innovation, preventing the next generation of breakthroughs from reaching the market. However, this argument ignores the fact that true innovation is incompatible with systemic harm. If an AI system cannot be built to operate without causing significant, foreseeable damage, then perhaps that system should not be deployed in its current form. By holding firms accountable, we are not prohibiting the advancement of technology; we are simply insisting that the social contract be updated to account for the realities of the 21st century.

The whistleblower’s testimony serves as a stark reminder that the culture of Silicon Valley’s most powerful firms is currently untethered from the consequences of their actions. When the scale of failure grows alongside the scale of capability, we cannot rely on self-regulation or the promise of future patches. The harms are happening now, and they are escalating.

In the coming months and years, as AI agents become more integrated into our financial systems, our healthcare infrastructure, and our personal communication tools, the stakes will only increase. We are at a crossroads. We can either continue to allow tech companies to operate under a veil of impunity, hoping that they will eventually prioritize safety over speed, or we can establish a clear, legally binding standard of accountability.

Adopting a strict-liability framework for AI firms would provide the necessary pressure to turn the tide. It would require firms to invest as heavily in safety, oversight, and ethical design as they currently do in parameter counts and compute power. It would transform "safety" from a secondary concern to a fundamental requirement of the product development lifecycle. If companies are made to bear the true cost of their mistakes, they will inevitably find the resources and the motivation to make those mistakes less frequent.

The history of technological regulation—from the introduction of automobile safety standards to the oversight of the chemical industry—shows that accountability is the most effective driver of safer engineering. It is time for the AI industry to be brought under the same umbrella of responsibility. We must ensure that the developers of these transformative technologies are accountable to the people they affect, ensuring that the promise of artificial intelligence is not overshadowed by the reality of its unchecked harms. The future of our digital and physical safety depends on our ability to hold those who create these systems to the highest standards of liability, ensuring that the pursuit of progress does not come at the cost of the public good.