Beyond Rudder Lube and Ozone Holes: A Pharmaceutical Governance Framework for Frontier AI (Consolidated Policy Brief)
This consolidated policy brief critiques the two dominant regulatory metaphors in contemporary frontier AI governance: commercial aviation (FAA/EASA deterministic engineering checklists) and the 1987 Montreal Protocol (global phase-out of fungible industrial chemicals with drop-in commercial substitutes), demonstrating why both frameworks are ontologically mismatched to adaptive, context-sensitive deployed AI systems. Drawing on modern biomedical and psychopharmacological regulatory history—anchored by FDA pharmacologist Dr. Frances Oldham Kelsey's 1960–1962 refusal of thalidomide, the resulting 1962 Kefauver–Harris Drug Amendments requiring empirical proof of safety and efficacy prior to mass release, and the contemporary Risk Evaluation and Mitigation Strategies (REMS) used for thalidomide derivatives (lenalidomide, pomalidomide) in hematological oncology—this brief translates proven pharmaceutical instruments into an actionable regulatory architecture organized around the Therapeutic Index (TI = TD50 / ED50) and operator skill floors.Specifically, the framework establishes: (1) a Five-Tier Controlled Deployment Scheduling Ladder (Schedule I Prohibited, Schedule II / AI-REMS Controlled Dual-Use, Prescription Rx, Behind-the-Counter BTC, and Over-the-Counter OTC) with an automatic schedule elevation trigger and a lateral IND/IRB Research Safe Harbor decoupled from hyperscaler NDA vetoes; (2) mandatory Phase IV post-market pharmacovigilance anchored to a public, liability-shielded Adverse AI Event Registry (AAER) with five severity grades and binding reporting clocks (24 hours to monthly); (3) standardized Compound AI Black-Box Warnings and compounding-pharmacy formulation/marketing liability for third-party wrapper vendors (with user-threshold exemptions for small open-source deployments); and (4) jurisdictional pluralism accommodating U.S. informed-consent and closed-loop progressive reality-drift titration alongside EU precautionary difficulty floors, local-compute technical friction gates, and the Shulgin Cautionary Principle (PiHKAL/TiHKAL) protecting independent investigators. Finally, the brief outlines three structural caveats where the pharmaceutical analogy bends—zero-marginal-cost open-weight replication (the BitTorrent problem), Big-Tech regulatory capture and First Amendment bounds, and continuous cloud-substrate mutation resembling living biologics rather than small-molecule tablets.
Authors
- Jesse Luke (ORCID: https://orcid.org/0009-0007-4059-9352)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24421095
- Primary Topic
- Ethics and Social Impacts of AI
- Type
- article
- Field-Weighted Citation Impact
- 0.00