Zendoric
← Back to the day · October 22, 2026

HHS extends Claude through a direct contract: government gets used to frontier AI before putting it out to tender

🕒 Published on Zendoric: October 22, 2026 · 00:19

✨ AI-generated · how it's made

The U.S. Department of Health wants to award Anthropic a no-bid contract to avoid cutting off access to Claude. It is a bridge to a larger tender, and it shows that AI is already infrastructure in government.

By ExecutiveGov · October 21, 2026.

The U.S. Department of Health and Human Services (HHS) plans to award Anthropic a sole-source contract, meaning one awarded without competition. According to a special notice posted Monday on SAM.gov, it would be a limited "continuity of services" contract. It would keep Claude for Enterprise and Claude for Government running for the department's authorized staff.

The contract would last until HHS awards a multiple-award blanket purchase agreement (BPA). This mechanism lets the government buy from several vendors under a common framework. According to the published information, that BPA will be the vehicle for accessing frontier AI models. The notice gives no award date, no contract value and no timeline for the BPA.

The immediate backdrop is the expiration, scheduled for October 31, of Anthropic's OneGov agreement with the General Services Administration (GSA). That deal gave agencies access to both products for one dollar. According to the outlet, HHS had already rolled out Claude to all its employees in December 2025.

The services covered are broad. They include chat, writing, summarization, research, reasoning and long-context models, document and data analysis, coding assistance, agentic features (tasks delegated to the AI), connectors and secure API access. HHS justifies the sole-source award by the need to keep using Claude's family of proprietary models. According to the notice, they would run with encryption, U.S. data residency, audit logs and handling of protected health information where applicable.

Our take: the story is small in money terms, since the value is unknown, but its method is revealing. A one-dollar price works as an adoption wedge. When it expires, the agency already has workflows built on one vendor, and switching is expensive. That is why the bridge contract appears. It is the same lock-in pattern we see in the cloud, now applied to language models.

This is not an accusation of wrongdoing. Agencies use these mechanisms legally to avoid disruption, and the subsequent BPA should open up competition. But follow-up matters. If the big procurement is delayed, the bridge could be extended. More broadly, vendor concentration in government is a market-power risk worth watching.

There is a more optimistic angle. A department that oversees public health, biomedical research and health programs is using AI at scale, with security and traceability requirements in writing. That is what it takes for AI to reach medicine with proper safeguards. In the short term there will be friction: costs returning after the bargain period, concerns about sensitive data and pressure on administrative jobs. In the long term, that same infrastructure is the path to faster, cheaper healthcare. The key is for public procurement to involve real competition and keep data under control.

🔗 Related on Zendoric

Sources & references