AnyBio today announced the availability of govern-js, the open-source SDK for AnyBio Govern, its governance layer for AI that communicates with patients. Published on npm as @anybio/govern, the SDK gives healthcare software teams one server-side call between the text their AI produced and the moment it reaches a patient. Each draft is checked against the organization’s own rules, held for a person when the rules require it, and recorded either way.

The question of who reviews what an AI tells a patient now arrives with the API key. Anthropic’s updated Usage Policy, effective November 12, 2026, requires that a qualified person review a high-risk medical recommendation before it is delivered to an individual, and that the individual be told AI was used. OpenAI’s Usage Policies, effective November 20, 2026, prohibit tailored advice that requires a license without a licensed professional’s involvement, and high-stakes medical actions without human review. State laws in force since January 2025 add disclosure and a path to a human. Whatever model an application runs, those duties land on the organization that deploys it: who reviews, and what it can show afterward.

“Healthcare teams need to explain why an AI message reached a patient, which rules applied, and when a person stepped in,” said Stephen Saine, founder and CEO of AnyBio. “govern-js makes that record part of the application, so reviewers read what happened instead of reconstructing it.”

AI is already inside the hospital record. A September 2025 data brief from the Office of the Assistant Secretary for Technology Policy found that 71 percent of responding U.S. non-federal acute care hospitals used predictive AI integrated with their electronic health records in 2024, up from 66 percent in 2023. [1] AnyBio described the design behind govern-js in its October 19, 2026 comment on the U.S. Food and Drug Administration’s discussion paper on generative AI-enabled medical devices. [2] The comment recommended evaluating the deployed configuration, including its policy controls and human review, and recording the model version, policy version and decision for each output.

How it works

The application sends each proposed patient message to the AnyBio Govern service before delivery. With enforcement on, the service returns one of three decisions.

Allow. The message passes the configured policy checks and proceeds to delivery.

Hold for a person. The message is withheld and routed to a reviewer the organization has named. If the patient is waiting on a reply, the application can send a holding line the clinical owner wrote in advance.

Block. The message is stopped, including when the patient has opted out or never consented on that channel, and the attempt is recorded.

The record ties the draft to the policy version, the decision and its reasons, any reviewer release and the released text. For delivery confirmation, the application computes a SHA-256 hash of the text it reports as sent and submits the hash, which the organization can check against its own copy. The organization holds the record, in one form across every model and application it runs, and clinical, compliance and engineering teams inspect the same record.

Observe first, then enforce

Every policy can run in observe mode before it enforces. Observe mode evaluates each message and records what the gate would have decided, without changing anything patients receive. The resulting report shows how many messages a week would have been held, and why, so review staffing rests on a number the organization has seen. For organizations already running patient-facing AI, AnyBio offers two weeks in observe mode followed by a report that reviewers can sign off on, so enforcement is turned on policy by policy, on evidence from the organization’s own traffic.

Developers work inside their existing stack

govern-js ships adapters for the Vercel AI SDK and LangChain, and a raw client for everything else. The command npx @anybio/govern init detects the application’s stack and prints the install steps. The command npx @anybio/govern test-message posts a synthetic draft and prints the decision, so a developer watches the gate hold a message with no patient data involved. Source code is at github.com/anybio/govern-js under the Apache-2.0 license, and each npm release carries a provenance statement.

The AnyBio Govern node for n8n calls the same Govern API directly. This verified community node runs on n8n Cloud and self-hosted n8n. Teams place it between the step that drafts a message and the step that sends it. The workflow can deliver the draft, show a holding line the clinical owner wrote in advance, or send nothing. A trigger node receives signed webhooks when a reviewer releases or rejects a held message or when a hand-off changes state.

Who it is for

govern-js is for teams building or operating AI that communicates with patients, including patient engagement, care navigation and between-visits care programs. The organization defines the policies and names the people who review. The SDK connects those decisions to the application. The hosted AnyBio Govern service supplies the checks, the review workflow and the record, and operates under a SOC 2 Type II report across all five trust services criteria and HIPAA, with model providers under business associate agreements and zero data retention. Organizations remain responsible for clinical validation, appropriate staffing and oversight, and for the obligations that apply to their products and workflows.

Developer documentation, including the n8n guide, is at developer.anybio.io. Organizations running patient-facing AI today can request the observe-mode review at anybio.io/ai-governance.

About AnyBio

AnyBio builds infrastructure for health software that runs on patient-generated data. It connects device signal capture, governed AI workflows and delivery into clinical systems, with provenance that follows the data. AnyBio builds for health product teams, care platforms, providers and research organizations.

[1] ASTP Data Brief No. 80, September 2025. Hospital trends in predictive AI, 2023-2024

[2] FDA docket FDA-2026-N-7874, discussion paper on generative AI-enabled medical devices. Public comment from Any Biosignal, Inc., October 19, 2026

Media gallery

About The Author
Business Wire