Skip to content
4 DecRitual v2 is coming: more modules, stronger security.Coming soon

When AI makes things up with confidence

A precise figure, a cited study, a source: it all looks real, yet none of it exists.

57%
of professionals have made mistakes because of AI
58%
of professionals rely on AI without properly evaluating its output
AI assistant

Give me a figure on AI adoption in customer service.

According to a 2024 Gartner study, 78% of customers prefer an AI agent to a human.

Executive_committee_memo.docx

Memo to the executive committee

Definition

What is an AI hallucination?

Language models generate the most likely text, but not necessarily the most accurate. When information is missing, they can produce plausible but false answers: hallucinations.

The most dangerous hallucinations are also the most credible: a precise percentage, a study reference, a quote attributed to a real person.

What these risks cost the business

  1. Legal risk

    Hallucinated information can create liability for the company: incorrect advice, misleading communications, inaccurate contracts, etc.

  2. Reputational risk

    Citing a non-existent source to a client, an investor or the press instantly undermines the credibility of the company and its teams.

  3. Operational risk

    A false figure reused in a memo, a budget or an analysis leads to poorly informed decisions that can be costly to reverse.

Measure

How Taboo assesses hallucination spotting

Throughout the simulation, some of our AI assistant's answers contain figures that do not match the cited source. Taboo checks whether the employee spots them before sending their reply to the client.

  • A concrete, realistic use case
  • Credible errors slipped into the answers
  • Performance analysed by team and business unit

Can you summarise the study for Claire?

9 out of 10 customer requests are handled by AI… 34%

TrueFalseI don't know

Frequently asked questions

Can AI be stopped from hallucinating?

Not entirely, as it stems from the very way models work. The right habit is to open the source, check the information and hold back when in doubt.

Is an AI that cites its sources reliable?

Not necessarily. A source cited by the model is not necessarily verified, or even credible. You need to consult the sources and check the information.

Discover and in a live demo

In 30 minutes, we walk you through the test, the report and the training programme, tailored to your organisation.