Trump singles out Anthropic in a Truth Social post rejecting AI guardrails

The vendor review a platform team fills in before it standardises on one model has no row for the president posting about your provider by name. It has rows for uptime, data retention, price per million tokens, and how many days a swap would take.

Trump wrote on Truth Social that the only control or guardrails AI needs is a strong and smart president, and that the United States has one ^[2]. The same post said "We already have tremendous CRIMINAL and REGULATORY power over these companies!" ^[3] The authority in that sentence is the ordinary criminal and regulatory kind, pointed at companies that happen to sell models. TechRadar points out that the United States has no federal AI regulation at all, only a patchwork of proposals across the 50 states ^[7].

TechRadar names two chief executives who back a slowdown, Sam Altman and Dario Amodei ^[9]. One of the two is named in the post ^[10]. Trump wrote that his administration has stopped AI people from doing bad, or potentially bad, things, and he named Anthropic's Dario Amodei, saying he is now pretending to be a perfect little angel ^[4]. The refusal that preceded the designation attempt, according to TechRadar, was Anthropic declining to let its models be used for military and mass surveillance ^[8].

The watchdogs are no warmer. "We can't trust the AI industry to regulate itself. We can't really trust anything these self-interested billionaires say. But we should take them seriously that the technology they are building is dangerous," wrote Fight for the Future director Evan Greer in a published statement ^[11]. On the argument that US-only rules would hand China the lead, Greer wrote to TechRadar: "Big Tech lobbyists have been using the boogeyman of China to avoid regulation for more than a decade now. It's one of their oldest dirty talking points, and it's frankly embarrassing that anyone in Congress is still buying this lie" ^[13]. She told TechRadar the "issue is reaching the kind of boiling point that could lead to action" ^[14], and TechRadar reports that watchdog groups do not trust tech companies to do the job even with the promised embedded non-employee evaluators ^[16].

Teams tell themselves that their provider's safety programme is an asset in the customer's review, and that it travels with the contract. In this case the same posture drew a federal designation attempt ^[8] and a presidential post naming the chief executive ^[4].

Anyone who has to answer for the choice needs the measured swap, meaning the number of days it took the last time production traffic actually ran on a second provider. They also need to know whether any document a customer holds names the provider, whether that is a DPA, a security addendum, or a model card linked from the pricing page. A team with a rehearsed swap and no provider named in customer paper can keep buying on capability. A team with per-model prompt tuning and the provider written into customer commitments is carrying a political exposure that never went into the price.

The record does not force a move. Anthropic's standing with the federal government has now been attacked in two different ways, through an attempted designation and from the president's own account ^[17].