Schiff Backs a Dedicated Federal Agency for AI Oversight
Sen. Adam Schiff, D-Calif., said he supports creating a new federal agency to regulate artificial intelligence in an interview published Oct. 5, 2026, on The Verge's Decoder podcast. The Verge The episode is titled "California Sen. Adam Schiff on AI, regulation, and corruption."
Schiff is a Democratic U.S. Senator from California. He sits on committees with oversight of intellectual property, antitrust, privacy and technology.
In the interview, Schiff discussed a formal regulatory framework for AI and compared AI regulation to social media regulation. The episode includes discussion of the tech backlash.
Intellectual property was also part of his portfolio in the conversation. His Senate office describes his committee work as advocating for intellectual property rights relating to patent, copyright, AI, and trademark issues. Senate office
The interview extended beyond technology policy. It covered what Democrats might do if they retake Congress in the midterms and whether Schiff would support impeaching President Trump over corruption.
The broader context here is institutional design. For people who build and run AI systems, the choice between a new agency and shared oversight by existing regulators is not minor detail. It decides where testing expertise sits. It shapes how rules for audits, record-keeping and pre-release checks are written, and how fast those rules can change when the technology shifts. A standalone agency would mean dedicated staff for model testing, incident reporting and standards work. Shared oversight would mean coordination across intellectual property, competition and privacy regulators, each using its own legal powers.
In my view, the comparison to social media regulation deserves close attention from technologists. Social platforms were regulated after they had already scaled, after their data interfaces, recommendation software and content-moderation systems were firmly in place. AI rules are being debated while the core pieces are still changing: large base models trained on broad data, fine-tuning methods that adapt them for specific uses, the server systems that run them, and agent software that can use tools on a user's behalf. That early timing cuts both ways. It gives policymakers a chance to set expectations for data origins, testing and public disclosure sooner. It also raises the risk of locking in compliance rules that fit poorly with how models are actually trained, tested and shipped.
For builders, one detail is worth flagging. Patent, copyright, trademark and AI-specific questions touch different parts of the same system. Taking in training data raises copyright questions. Model behavior and outputs raise separate questions about ownership and liability. Names, faces and brands used in prompts or outputs raise trademark and publicity questions. Engineers often see one data pipeline. The law sees separate regimes. Any unified framework will need clear definitions, clear standards of proof, and a clear map of responsibility across data providers, base-model developers, deployers and end users.
On the practical side, staffing will decide much. Effective oversight needs people who understand testing setups, the limits of adversarial testing, where training data comes from, speed tradeoffs in running models, and how systems fail in production. Without that in-house expertise, rules tend to become paperwork. With it, regulation has at least a chance to improve system quality, clarify rights for creators and give businesses a stable basis for investment. That outcome is not guaranteed, but it is the version worth working toward.
