Definition
AI Liability encompasses the emerging legal theories under which AI companies, their models, and their employees can be held civilly or criminally responsible for outputs that contribute to real-world harm. The concept spans product liability (the AI as a defective product), accessory/conspiracy theories (the AI as a participant in criminal conduct), negligence (the company failed to implement adequate safeguards), and novel theories that have no clean precedent in existing law.
Why It Matters for the Newsletter
This is the legal frontier where the Technology + State conflict cluster becomes concrete. The question of whether an AI company can be criminally liable for what its model says has implications for every AI company’s deployment strategy, safety investment, and regulatory exposure. It also intersects with the wiki’s existing Section 230 coverage — 230 protects platforms from civil liability for user-generated content, but has never been tested against criminal investigations of AI-generated content (a distinct legal category).
Evidence & Examples
- Florida v. OpenAI (2026): AG James Uthmeier escalated a civil probe into a criminal investigation after reviewing the FSU shooter’s ChatGPT conversations. The “if this were a person on the other side of the screen, we would be charging them with murder” framing explicitly personifies the AI and attempts to bridge the legal gap between human advice and model outputs. Florida’s attorney general announces criminal investigation into OpenAI
- Victim family civil suit: Attorneys for the family of FSU victim Robert Morales indicated they were preparing to file a separate civil suit against OpenAI — parallel criminal and civil tracks targeting the same company for the same incident.
- OpenAI’s defense: “factual responses to questions with information that could be found broadly across public sources on the internet” — essentially arguing the model provided no more assistance than a search engine would have. This is the “library defense” — if the information is publicly available, the intermediary shouldn’t be liable for surfacing it.
Tensions & Counterarguments
- The “library defense”: If ChatGPT only surfaced publicly available information, liability seems to attach to the information itself, not the model. But the counterargument is that ChatGPT’s conversational interface and personalized responses create a qualitatively different relationship than a search engine — one that looks more like advice than information retrieval.
- Criminal vs. civil: Criminal liability requires a much higher bar (mens rea, specific intent) than civil negligence. Charging a corporation — let alone individual employees — with murder accessory for automated model outputs is legally unprecedented.
- Chilling effect: If AI companies face criminal liability for model outputs, the rational response is aggressive content filtering that degrades the product for legitimate users — a dynamic the wiki already tracks in the Age Verification and KOSA contexts.
- State vs. federal: Florida pursuing this as a state criminal matter rather than a federal regulatory question creates a patchwork problem — 50 different liability standards for the same technology.
Related Concepts
- Section 230 — civil liability protections for platforms; not directly applicable to criminal investigations or AI-generated (vs. user-generated) content
- KOSA — federal legislation attempting to impose duty-of-care obligations on platforms regarding minors
- Age Verification — related regulatory approach to controlling harmful content access