HHS plans a no-bid award to keep Claude running for its workforce

What HHS posted

The US Department of Health and Human Services wants to keep using Claude without holding a competition first. In a special notice posted on SAM.gov on 19 October 2026, titled "Intent to Sole Source to Anthropic," the department said it plans to solicit, negotiate and award a contract to Anthropic as the only source for continued enterprise access to its Claude models.

The notice, number 7571TE27R00001, was posted by the information technology office of the OMAS Strategic Buying Center, which sits within HHS's Office of the Assistant Secretary for Financial Resources. It lists Rockville, Maryland as the place of performance. It is explicit that it is not a request for competitive proposals; companies that believe they could meet the requirement had until 3 p.m. Eastern time on 21 October to respond.

HHS describes the purchase as "a limited continuity of services award" that will run while it prepares a broader competition. FedScoop, which first reported the notice, said neither HHS nor Anthropic answered its questions about it. The notice gives no dollar value and no length for the award.

What the award would cover

The award covers continued enterprise access to "Claude for Enterprise / Claude for Government" for authorised HHS staff and other users. The notice lists the capabilities in some detail: general chat, reasoning, drafting and summarisation, search and research, file and document workflows, and reporting and analytics, along with the highest-capability reasoning models and long-context modes.

It also names newer, more agentic features. HHS wants coding assistance and developer tools, agentic and delegated-work features, connectors, projects and workspaces, memory and personalisation "where authorized," web-grounded research where authorised, native applications, and secure API or gateway access. On the administrative side, it lists enterprise single sign-on, SCIM user provisioning, role-based access control, group and feature controls, reporting and audit export.

That list reads as a description of what HHS staff already have rather than a new programme. The notice says the requirement is continuity of "the existing Claude deployment already integrated into HHS infrastructure," and refers to a Section 4 that is not included in the public posting.

Why HHS says only Claude will do

The legal basis is the part procurement experts noticed. HHS cites 41 U.S.C. 1901 and Federal Acquisition Regulation 13.106-1(b)(1), the simplified acquisition rules that let a contracting officer solicit from a single source when only one is reasonably available. The notice frames this as "Only One Responsible Source and No Other Supplies or Services Will Satisfy Agency Requirements."

Its justification is short: "The capability is specific to one provider: Claude is Anthropic's proprietary model family, and the requirement is continuity of the existing Claude deployment." In other words, HHS is not arguing that no other company sells a capable AI assistant. It is arguing that switching away from the one it has deployed is not an option for the interim period.

Jessica Tillipman, associate dean for government procurement law studies at George Washington University, told FedScoop that the notice reads like an admission that the department depends on Claude. She said the capabilities described are fairly general chatbot services, that claiming only Claude can provide them is a questionable use of sole-sourcing, and that if switching is too painful or costly, the department negotiated poorly despite warnings about vendor lock-in.

The security and data terms

The notice spells out the data-handling terms the deployment must meet, which give an unusually concrete picture of what a large civilian agency now requires from an AI vendor. HHS data must be excluded from model training by default. Encryption must be FIPS 140-validated in transit and at rest, and data must stay in the continental United States.

HHS also requires a HIPAA business associate agreement posture for surfaces eligible to handle protected health information, exportable logs and audit trails, notification of security incidents within one hour, retrieval of data within one business day on request, and secure deletion to the NIST SP 800-88 standard with export in a non-proprietary format when the contract ends.

Those exit terms matter given the lock-in criticism: they are the conditions that would make a later switch possible. They also echo the direction of federal buying rules more broadly, such as GSA's new clause on safeguarding government data in large language model systems, which agencies must start using by 19 October, the day the HHS notice went up.

How HHS got here

According to FedScoop, HHS rolled Claude out to its workforce in December 2025, after Anthropic offered its services to federal agencies for $1 a year under the General Services Administration's OneGov programme, and listed uses of Claude in its 2025 AI inventory. FedScoop reports that the discounted OneGov deal, already an extension of the original, ends on 31 October, which would explain the timing of a stopgap award.

The relationship survived a sharp break between Anthropic and the Trump administration. The Pentagon designated Anthropic a supply-chain risk this year, a decision Anthropic is now asking the full D.C. Circuit to rehear. FedScoop reports that the dispute led to a governmentwide ban on Anthropic and that the Food and Drug Administration, an HHS agency, began moving its Elsa chatbot from Claude to Google's Gemini earlier this year. The HHS notice suggests that, at least outside defence, parts of government still rely on Claude day to day.

HHS is also trying other vendors. It is one of the early users of Google's Gemini for government services preview, which launched on 20 October. The department appears to be running more than one AI assistant at once while it decides how to buy them long term.

The competition that comes next

The sole-source award is meant to bridge to a competitive purchase that HHS calls the "HHS Enterprise Frontier SI Models and Advanced Features multiple-award BPA." A blanket purchase agreement with multiple awardees would let HHS buy from several AI vendors under pre-agreed terms and place orders with whichever suits a given use.

FedScoop said it was not clear whether that competition had started. A SAM.gov search for the BPA's name on 21 October returned only the Anthropic notice. If and when it is issued, it will show whether HHS intends to keep Claude as one of several options or to use the competition to reduce its dependence on a single supplier.

The broader federal market is moving to paid terms at the same time. FedScoop notes that the end of the OneGov discounts has prompted at least OpenAI to announce consumption-based pricing for agencies, and the HHS notice does not say whether the bridge award will be priced per seat, by usage or as a flat fee.

Questions the notice leaves open

Several basics are missing from the public notice. It does not say how much the award will cost, how many HHS staff have Claude accounts, how long the bridge will last, or whether it covers the whole department or only some of its agencies. Nor does the posting include the Section 4 it cites on the existing integration with HHS systems.

It also does not address the political context. The administration has been tightening rules on which AI tools agencies may use; just this week DHS ordered federal civilian agencies to remove DeepSeek. The HHS notice treats Anthropic as an ordinary, if dominant, supplier, and contains nothing about the Pentagon dispute.

Why it matters

Government AI deals so far have been sold on near-zero prices. The HHS notice is an early look at what happens when those introductory offers expire: an agency that has woven one model into daily work concludes it cannot switch in the short term and turns to a no-bid award to keep it running.

For Anthropic, it is a sign that a large civilian department still wants its product despite the Pentagon fight. For procurement watchers, it is the vendor lock-in risk they warned about, written into a federal notice. The coming multiple-award competition will decide whether that dependence becomes permanent or is the price of a smooth transition.