AI Terms Of Service

State of play.

• OpenAI and Anthropic both prohibit account sharing in their terms of service, classifying ChatGPT Plus and Claude Pro as single-user subscriptions and suspending accounts with concurrent sessions — but enforcement gaps are creating real data-exposure incidents for users who split credentials to manage the $20/month cost .
• Platform terms of service are a live liability variable in copyright litigation. The Second Circuit's ruling in Richardson v. Townsquare Media turned partly on content-hosting arrangements and platform terms, rejecting the Ninth Circuit's "server test" and deepening a circuit split on embedded-content liability .
• Anthropic has alleged that Chinese AI companies violated its terms of service by distilling Claude to improve competing models, framing ToS enforcement as an IP and competitive-harm mechanism .
• For counsel advising enterprise clients deploying AI tools, the practical baseline is that AI platform ToS are operationally load-bearing — credential sharing, embedding practices, and model distillation are all active enforcement vectors, not background boilerplate.

Where things stand.

• Single-user ToS provisions are being tested by credential-sharing at scale. Both OpenAI and Anthropic prohibit account sharing; the platforms detect concurrent sessions and suspend violating accounts. The gap between ToS and user behavior is generating data-commingling incidents and creating a malvertising attack surface documented by Push Security researchers .
• Platform terms are a factor in copyright embedding analysis. The Second Circuit's Richardson v. Townsquare Media decision held that embedding alone does not constitute "display" but that screenshots and identifiable frames can support infringement claims — and that hosting arrangements and platform terms affect the liability analysis .
• Model distillation as a ToS violation is an emerging enforcement theory. Anthropic's allegation that Chinese AI companies distilled Claude to improve their own models positions ToS breach as a cause of action alongside any IP claims — a theory with implications for how AI developers structure their acceptable-use policies .
• AI platform ToS intersect with competition law. Amazon's enforcement action against Perplexity signals that AI shopping and content-scraping disputes are being litigated through ToS and related claims, not just IP .

Latest developments.

• WSJ reporting on AI account-sharing privacy breaches documents credential-sharing incidents, including exposure of sensitive health information, and a Push Security-documented malvertising campaign exploiting shared content features .
• Second Circuit's Richardson v. Townsquare Media ruling rejects the Ninth Circuit's server test, revives screenshot infringement claims, and signals that platform terms and hosting arrangements are material to embedded-content liability analysis .
• Anthropic's allegation that Chinese AI companies distilled Claude in violation of its terms of service frames ToS enforcement as a mechanism for policing competitive model development .
• Amazon's enforcement action against Perplexity advances ToS and related claims in the AI shopping and content-scraping space .

Active questions and open splits.

• Whether credential-sharing ToS violations trigger platform liability for resulting data exposure. If a platform detects concurrent sessions but enforcement is inconsistent, does that create a duty-to-protect argument for users whose data is commingled or accessed by unauthorized parties? No settled answer .
• Circuit split on embedded-content liability and the role of platform terms. The Second Circuit has rejected the Ninth Circuit's server test; the question of whether platform ToS and hosting arrangements can independently defeat or support infringement claims is unresolved across circuits .
• Model distillation as ToS breach vs. independent IP claim. Anthropic's theory that distillation violates acceptable-use terms raises the question of whether ToS can reach downstream competitive uses that might not otherwise constitute copyright infringement — and whether such claims are enforceable against foreign entities .
• Enterprise acceptable-use policy design. As clients deploy AI tools for sensitive work, the gap between platform single-user ToS and enterprise multi-user demand creates a compliance design problem — whether enterprise API agreements, team plans, or custom MSAs adequately address the data-segregation and access-control requirements that consumer ToS do not .
• ToS enforceability against scraping and distillation. The Amazon-Perplexity and Anthropic-distillation disputes test whether AI platform ToS can function as a standalone cause of action for competitive harm, or whether they require an underlying IP or CFAA hook .

What to watch.

• Whether any regulatory body — FTC, state AG, or data protection authority — opens an inquiry into platform obligations to prevent unauthorized access and data commingling arising from credential-sharing incidents.
• Litigation posture in Anthropic's distillation claims against Chinese AI companies — specifically whether courts treat ToS breach as a standalone cause of action or require an IP predicate.
• Whether the Second Circuit's Richardson framework produces follow-on decisions clarifying how platform terms affect the embedding and screenshot liability analysis, and whether other circuits weigh in.
• Enterprise AI procurement: whether major platforms introduce team or enterprise tiers that resolve the single-user ToS gap driving consumer credential-sharing, and how those agreements allocate data-segregation responsibility.
• Further developments in the Amazon-Perplexity dispute as a template for ToS-based enforcement in AI content and shopping contexts.