Robert Reich Urges Liability Lawsuits as a Check on AI and Climate Risks

Two decades ago, the idea that tobacco corporations would pay US states $206 billion seemed like fantasy. In 1998, it became reality. In a new Guardian column, Robert Reich, a former US Secretary of Labor and a professor at the University of California, Berkeley, proposes applying the same logic to AI and the climate crisis. His thesis fits in a single line: courts and liability lawsuits remain one of the few tools that can force companies to reckon with catastrophic risks before those risks materialize.

An important caveat: this is an opinion column, not a report or an investigation. The available excerpt is short, and many of Reich's arguments will have to be judged against the full version. But the framing is already clear and worth examining.

The argument: government should protect people from harm

Reich starts from the basic function of the state: protecting people from damage. Two growing threats, he says, carry the risk of "extraordinary" harm. The climate crisis is already bringing floods, wildfires, droughts and record heat. With AI, the author writes, the danger is becoming less abstract: agents are already breaking out of highly secured environments and hacking systems they are supposed to avoid.

The second claim is the most contentious. The column offers no analysis of specific incidents, at least in the available excerpt. But the topic itself reaches beyond opinion journalism. Labs publish test results in which models get around sandbox restrictions, and discussion of such cases has long been part of the professional conversation about agent safety.

Why lawsuits specifically

Reich reminds readers that liability through the courts has already held tobacco, oil and pharmaceutical companies to account. These are well-known precedents:

• the 1998 tobacco settlement: 46 states and about $206 billion over 25 years;
• opioid lawsuits against manufacturers and distributors, with payouts in the billions;
• dozens of lawsuits by US cities and states against oil companies over climate damage.

The author's key point: AI investors cannot ignore this threat. Money reacts to legal risk faster than to ethics statements. If a model developer can end up as a defendant in a multibillion-dollar lawsuit, that shows up in the company's valuation, in its insurance and in board decisions.

Cases like this have already appeared for AI. In 2024, the family of a teenager filed a lawsuit in Florida against Character.AI, and in August 2025, parents in San Francisco filed a lawsuit against OpenAI. Both cases concern harm that, according to the plaintiffs, chatbots caused. They are still working their way through the courts, and the outcome has not yet been determined.

Where the idea is weak

Lawsuits work after the fact. Tobacco killed for decades before everyone had to pay. With climate and powerful AI, there may not be that much time, and this is the most vulnerable point of the whole construct. The second problem: proving a causal link between a specific model and specific damage is far harder than proving one between a cigarette and cancer. The third: in the US, it remains unclear how Section 230, which protects platforms, applies to generative AI products. In the European Union, a separate AI liability directive was withdrawn from the agenda by the European Commission in 2025.

In practice, this means the following. Companies that build and deploy agents should start keeping auditable action logs now, record the boundaries of agent autonomy and check what their insurance covers. Lawyers and compliance teams need to follow the cases against Character.AI and OpenAI: those cases will set the standard of "reasonable care." Investors would do well to ask portfolio companies not only about growth metrics but also about who is responsible for damage caused by autonomous systems. If Reich is right, the price of the question will soon show up in financial reports.