When your company puts AI into products or customer service, legal's job flips: you now draft the terms customers sign, and the company carries the risk of what its AI says and does.
What customer terms should cover
- Clear description: what the AI feature does, what it doesn't, and its known limits.
- Customer responsibilities: reviewing output before relying on it, lawful use, the right to upload what they upload.
- Acceptable use: prohibited uses (for example, unlawful, high-risk or deceptive uses), with the right to suspend.
- Data use: whether customer data trains your models. Customers increasingly expect "no" by default, mirroring what you ask of your own vendors.
- Output ownership and IP: who owns outputs; whether you offer an IP indemnity, and on what conditions.
- Liability: caps, exclusions and carve-outs for AI-specific risks.
- Regulatory allocation: which party does what under AI laws (for example, provider vs deployer duties).
Disclaimers have limits
A disclaimer that "AI may make mistakes" sets expectations, but it doesn't make the company immune. Consumer protection law, misrepresentation and sector rules still apply to what your AI tells people. Pair disclaimers with quality controls: testing, monitoring, escalation to humans and a fast correction process.
Claims about your AI
What sales and marketing say about the AI is a legal risk too. Regulators have acted against overstated AI claims, including the FTC's action against a "robot lawyer" service in 2024 and the SEC's 2024 "AI washing" cases against investment advisers.