AI hallucination (in support)
A hallucination is when an AI confidently states something that is not true: an invented policy, a made-up price, a step that does not exist. In customer support it is the single most damaging failure, and it carries real legal weight.
What it looks like in the wild
In 2025 a coding tool's support bot invented a nonexistent one-device policy and triggered real cancellations. In a 2024 tribunal ruling, an airline was held liable after its chatbot made up a bereavement refund policy. The lesson from both: you own every word your bot says. A confident wrong answer is worse than no answer.
Why it happens
Most support hallucination is not random invention. It is unfaithfulness to retrieved context: if the most relevant article is wrong or missing, the model delivers a wrong answer with the same confidence as a correct one. Neural networks are also systematically miscalibrated, so "seems sure" is not the same as "is right."
How Resolve prevents it
Grounding is the core defense. Reva only states facts it retrieved from your real content or live data, and if it cannot ground a claim it refuses and escalates instead of guessing. A separate supervisor model verifies every draft before it ships, business logic runs in code, and a forbidden-action list keeps the AI away from refunds, chargebacks, and legal answers. Refusing to guess is a rewarded outcome.
Frequently asked
How do you stop an AI support agent from hallucinating?
Ground or refuse: only state retrieved facts, compute policy in code, verify with a supervisor model, and route anything unsupported to a human. Resolve does all four and treats refusing to guess as a good outcome rather than a failure.
Can a company be liable for its chatbot's answers?
Yes. A 2024 tribunal held an airline liable for a refund policy its chatbot invented. You own every word your bot says, which is why grounding and a supervisor check matter.