SuperML
← All issues

FDA Has No Framework for Agentic Clinical AI. ARPA-H Is About to Create One.

ARPA-H is selecting teams to build the first FDA-authorized agentic clinical AI — a prescription-writing cardiovascular agent with no existing FDA validation framework, no clearance precedent, and a 39-month timeline that assumes regulators will solve governance problems nobody has solved yet.

Hi there,

The FDA has authorized over 1,350 AI medical devices. Every single one is a predictive, fixed-function tool — a model that takes inputs and returns a recommendation a clinician can review and override. ARPA-H is now selecting teams to build something fundamentally different: an autonomous agent that writes prescriptions, modifies care plans, and operates as a 24/7 cardiovascular care team member. The FDA has no framework for this. The governance problem is real, and it's not getting solved in 39 months without serious architectural investment.

🔥 Featured Post

FDA Has No Framework for Agentic Clinical AI. ARPA-H Is About to Create One.

  • ARPA-H ADVOCATE is selecting teams in June 2026 to build the first FDA-authorized agentic clinical AI for cardiovascular care — connecting to EHRs, writing prescriptions, and acting autonomously around the clock
  • FDA's January 2026 CDS guidance loosened oversight for recommendation AI — but only where clinicians can "independently review the basis for the recommendation." Autonomous prescription-writing agents cannot satisfy this condition
  • FDA's SaMD lifecycle guidance was designed for adaptive, continuously-learning predictive algorithms — not agents that chain diagnosis, medication adjustment, scheduling, and patient communication into a single autonomous workflow
  • ADVOCATE requires a supervisory agent to ensure the clinical AI's "consistent safety and effectiveness" — a meta-governance architecture that banking regulators just started wrestling with, and that FDA has never evaluated
  • Healthcare AI architects deploying clinical decision support today need to understand that the governance gap ADVOCATE will expose is already present in the systems they're shipping

Read the full post →

📚 In Case You Missed It

FinCEN AML: 'Effective' Means AI Now. Nobody Built the Governance Yet. — FinCEN's effectiveness-based AML standard implicitly rewards AI adoption — but banks racing to deploy ML for compliance credit have no model governance framework, and OCC examiners are asking about it in every exam.

Cursor Is Now SpaceX: Enterprise Agentic Coding's New Lock-In Risk — SpaceX's $60B Cursor acquisition ends model-neutral AI coding — enterprise teams built on Cursor's multi-model architecture now face a silent model substitution event in their agentic CI/CD pipelines.

OpenAI's Pre-Release Safety Trick: Make Models Think They're in Production — OpenAI replays 1.3M anonymized production conversations with candidate models before release — catching reward hacking that adversarial evals miss, with 1.5x median error in predicting undesired behavior rates.

More posts dropping every week. Stay sharp.

— Bhanu @ superml.dev