Why the Big Labs Are Slowing Down: It's Not Safety, It's Liability Picture a Hugging Face-style incident with Pearl Harbor-level impact: a model or weights released at scale that ends up enabling a coordinated attack on critical infrastructure. The legal fallout would not be one lawsuit. It would be every jurisdiction at once. Civil exposure comes first. Product liability and negligence claims from every affected party, consolidated into mass-tort litigation. Shareholder suits arguing the board knew the risk and shipped anyway. Discovery would surface every internal safety memo, every red-team result, every "ship it" email, and those documents become Exhibit A. Regulatory exposure stacks on top. The EU AI Act already allows fines up to 7 percent of global revenue. Sector regulators for energy, finance, and telecom carry their own penalties. Multiply that across the US, EU, UK, and Asia and the numbers stop being fines and start being existential. Then personal exposure. Directors have a duty of oversight. Executives who sign off on a release despite documented, unmitigated catastrophic risk are not shielded by the corporate veil. Criminal negligence and reckless endangerment statutes exist for exactly this pattern: you knew, you had the means to prevent it, you proceeded. A CEO facing that record is not thinking about benchmarks. That's why legal departments run these organizations now. No general counsel will let a company be the named defendant in the first AI-enabled mass casualty case, and no CEO wants to be the first executive deposed in it. We are too focused on lawyering. That hyper-focus is exactly why we will lose to China. Beijing is not slowing its labs down with the threat of trillion dollar judgments and decade-long litigation. We are. When the loudest voices in the room are general counsel and the plaintiffs' bar, and the incentive is a massive payday for lawyers rather than a working system for engineers, the outcome is predictable: engineering waits while lawyering wins. A country that treats every model release as a future deposition instead of a technical milestone is not racing. It is litigating. And if we stay a litigious country first and an engineering country second, we will not have lost the AI race in a lab. We will have lost it in a courtroom. The big AI labs are not slowing down because of safety. They are slowing down because once a global security attack traces back to your model, prosecution isn't a risk. It's a certainty.