Federal Agencies Deploy Medical AI for Therapy and Prescriptions Amid Safety Debate
The rapid federal rollout of AI systems that interact directly with patients creates a regulatory gap that existing Medicare and FDA frameworks were not designed to address.
Federal health agencies are moving to deploy artificial intelligence systems capable of providing therapy sessions and prescribing medication to patients, according to reporting by The New York Times published September 14, 2026. The systems are being integrated into programs that touch Medicare beneficiaries, raising questions about liability, clinical oversight, and the financial interests of private technology investors involved in the projects.
The deployments represent an active federal policy decision, not a pilot or exploratory phase. AI agents, defined in this context as software systems that take autonomous actions such as generating prescriptions or conducting structured therapy conversations, are being introduced into care pathways that serve some of the most medically vulnerable Americans, including elderly Medicare enrollees.
The Food and Drug Administration has regulatory authority over software that qualifies as a medical device under 21 U.S.C. 321(h). Whether AI systems that prescribe medication or conduct therapy fall within that definition is a threshold legal question the agency has not publicly resolved for these specific deployments, according to The New York Times report. The document that would clarify the FDA's position is a formal guidance or enforcement discretion policy, which the agency has not yet published for this category of autonomous clinical AI.
Medicare coverage and reimbursement for AI-delivered services would be governed by rules issued by the Centers for Medicare and Medicaid Services. CMS has broad authority under 42 U.S.C. 1395 to define covered services. Whether CMS has issued a formal determination on reimbursement for AI-conducted therapy or AI-generated prescriptions is not specified in the available source material. A CMS program memorandum or final rule would be the public record that would establish that determination.
The involvement of venture capital investors in the federal projects is noted in the Times report as a concern raised by critics. Federal contractors and grantees with equity backing from private investors are subject to disclosure requirements, but the specific firms involved and their financial relationships with the federal programs are not named in the available source material. Lobbying Disclosure Act filings and USASpending.gov contract records would identify which private entities hold federal agreements related to these deployments and what amounts have been obligated.
Safety concerns cited in the report center on the autonomous nature of the AI systems. A human clinician who prescribes medication incorrectly is subject to state medical board oversight, malpractice liability, and in some cases federal sanctions. The liability framework that applies when an AI system generates a prescription that harms a patient has not been adjudicated in federal court and has not been addressed by statute as of the date of this publication. The Federal Tort Claims Act, 28 U.S.C. 1346, would govern liability for harm caused by AI systems operated directly by federal agencies, but the extent to which that coverage applies to contractor-operated systems used in federal programs is an unresolved question.
The mental health component of the deployment is governed by additional federal law. The Mental Health Parity and Addiction Equity Act requires that mental health benefits not be more restrictively covered than medical or surgical benefits. Whether an AI-delivered therapy session satisfies parity requirements for purposes of Medicare is not addressed in publicly available CMS guidance as of September 2026.
Congress has not passed legislation specifically governing the clinical use of autonomous AI in federal health programs. Several bills have been introduced in the 119th Congress addressing AI in healthcare contexts, but none has cleared committee as of the date of this publication. The Congressional Record would reflect any floor consideration or markup activity on such legislation.
The pace of federal deployment stands in contrast to the posture of some AI developers. As previously reported by The Congressional Times, OpenAI and Anthropic have issued internal safety warnings that place them at odds with elements of the current administration's approach to AI governance, a dynamic that is relevant to understanding the broader federal policy environment in which medical AI is being deployed.
What remains unknown is the identity of all federal programs currently using autonomous clinical AI, the names of the private contractors or grantees operating those systems, the total dollar value of federal obligations to those entities, and whether FDA has made any internal determination about the device status of the software involved. USASpending.gov contract search records, FDA dockets, and CMS program documentation would be the public records capable of answering those questions.