An Air Canada chatbot promised a user a refund in the event of death – even though the company had no such policy. The problem: the AI was hallucinating. Air Canada was even held responsible for this in court.
Such errors are not isolated cases, but symptoms of a lack of security in the development and integration of AI systems.
At ISTE, we deal with precisely such questions:
How do we build AI systems that are reliable, transparent and legally robust? How would we have prevented such a mistake?
- AI system architecture with responsibility
Our experts develop and test architectures that ground LLMs in secure knowledge sources (e.g. via RAGs or knowledge graphs). No hallucination – no liability gap. - Knowledge graphs & policy governance
We formally model company-specific knowledge – from reimbursement policies to exception processes – and integrate it into AI systems in a controlled manner. - “Safe by design” evaluation
With our interdisciplinary framework (technology + law + UX), we ensure that systems hold back in critical cases instead of fantasising. - Audits & certifications for trustworthy AI
We offer accompanying quality and security evaluations for chatbots, agents and generative AI – from prompt testing to fact consistency checking.
Our goal: AI should not impress – it should deliver what it promises.
If you want to use generative AI in your company – in a secure, controlled and liability-compliant manner – talk to us. The next Air Canada moment doesn’t have to happen.