HHS's 'bridge' contract with Anthropic: the problem isn't the vendor, it's a procurement process that doesn't hold up

By FedScoop · October 23, 2026.

The US Department of Health and Human Services (HHS) plans to award Anthropic a six-month “bridge contract” without competition. According to a spokesperson quoted by FedScoop, the goal is to maintain continuity of “vital” AI services when the company's discounted offer expires on October 31.

That discount came through OneGov, the General Services Administration (GSA) program that offered access to AI tools for $1 for one year. Anthropic had already extended that agreement by one month. According to the same report, OpenAI did announce a consumption-based pricing agreement through 2028 before the expiration. Anthropic has not yet announced anything equivalent.

HHS says it is seeking a long-term contract and wants to finalize it before the bridge ends. According to FedScoop, the spokesperson did not say whether there is a public solicitation for that contract or what exactly justifies the sole-source award.

The weakest point, according to the experts cited, is the justification. Jessica Tillipman of George Washington University describes it as a circular argument: the only one that can provide Claude services is Anthropic. In her view, HHS does not explain why only Claude meets requirements that, she says, describe generic chatbot functions. The new details, she says, “weaken” the agency's case.

She also questions the planning. The agreement's end date was public, so it is unclear why the agency did not prepare for the transition.

David Timm, a government contracts attorney at Burr and Forman, adds another argument. After the clash between the Pentagon and Anthropic, President Trump ordered an administration-wide ban on its tools, and HHS deactivated its products. A federal judge has since suspended that ban. For Timm, if HHS was able to “make do” back then, likely with other models, the continuity argument becomes harder. He believes uniqueness and continuity are “a tough sell” and that a challenge “would have a basis.” He cautions that he cannot assess its chances without the full record.

The amount remains undisclosed. The award was made under the simplified acquisition threshold, which allows a streamlined process below $350,000. Timm notes that it is unclear whether that was the correct route, because the notice does not include the cost. Above that threshold, a formal justification would be required, which the notice also does not include.

Our take: what matters here is not Anthropic. It is what the episode reveals about how governments buy AI. The $1 per year was an entry price, not a real price. When it expires, the buyer is left with a technical dependency and no budget in place. Broadly, this is the classic “lock-in” pattern, dependence on a single vendor, and here it plays out in fast-forward.

Natalie Buda Smith, chief AI officer at the Library of Congress, summed it up at an event cited by FedScoop: agencies buy AI as a capital expense, but it is an ongoing service, like utilities. If so, it must be budgeted and procured as such, with viable alternatives from day one.

There is also an awkward irony. If, as Timm suggests, HHS managed without Claude during the ban, that is good news about the substitutability of models, but it undermines the “vital continuity” excuse. It is an argument in favor of a competitive market, not a sole-source award. We do not know whether there is a technical requirement that justifies the choice, and HHS has not explained it.

In the short term, the risk is more reputational and procedural than one of capability: a possible challenge from a competitor, although Timm doubts OpenAI or Google would be interested if they already sell to HHS. In the long term, the lesson is useful for the broader thesis. AI that ends up easing the healthcare and administrative burden needs institutions capable of buying it well, with competition, transparency and exit options. Without that plumbing, the benefit stays in the press releases.

All of the above is based on FedScoop's reporting and the opinions of the experts it quotes. There is no firm legal conclusion: that will depend on the record and on any possible formal protest.