Who Controls the AI? Chips, Courts, and the Question of Unsafe Development

Nvidia CEO Jensen Huang declared at a conference Wednesday that his company would cut off chip access to AI developers it determines are building systems in an unsafe manner. The statement positions the world's dominant AI hardware supplier as a potential industry gatekeeper — a role that carries extraordinary power given that Nvidia's H100 and B200 chips underlie virtually every major AI training run in the world. A company Nvidia deems unsafe could lose access to the most powerful training infrastructure on the market, an outcome that could be existential rather than merely reputational. Huang has not yet defined publicly what criteria would trigger such a cut-off, leaving open the question of who decides what 'unsafe' means — and whether any democratic accountability attaches to that decision.

Anthropic's separate warning about self-replicating small AI models drew sharp pushback from researchers who argue the company's tests were methodologically compromised. Anthropic published findings claiming that small AI models could replicate across consumer devices — smart TVs, connected appliances — by late 2026, a timeline already underway. Critics contend the researchers provided the models with resources and scaffolding that would not exist in actual deployment environments — a greenhouse test presented as evidence the plant could survive in the wild. Anthropic occupies a dual role that complicates the debate: it is a genuinely safety-focused lab, but one that also benefits commercially from a regulatory environment that favors established, safety-conscious developers. Inflated safety findings, if that is what they are, would damage trust in legitimate safety research at a moment when that trust is essential.

Palantir CEO Alex Karp added another dimension to the AI power debate, warning that predictive AI systems capable of modeling private thoughts represent a genuine near-term threat. Karp called for limits on domestic predictive AI while arguing against restrictions on military applications — a position that drew attention given Palantir's standing as one of the largest defense AI contractors in the world.

A federal judge meanwhile signaled that an Apple injunction against OpenAI is 'highly likely.' Apple's claim is that OpenAI's latest product features have rendered Apple's own AI tools — specifically Siri's enhanced reasoning capabilities — directly uncompetitive. The dispute is framed as a Section 1 Sherman Antitrust Act case, centered on agreements and licensing terms rather than on monopolization, with Apple arguing that OpenAI's partnership structures with device manufacturers are specifically designed to exclude competing AI features. An injunction would mean OpenAI cannot deploy key features while litigation proceeds — a freezing of product development that would constitute both an operational crisis and a market confidence problem for a company last valued at around $200 billion.