Palen AI reliability
How Palen reduces hallucination risk in its answers
Palen is not a general-purpose chatbot. Its ADR engine is surrounded by a state-of-the-art software architecture that constrains its sources, governs its analysis and controls the traceability of its conclusions. With no Web access, the AI remains inside this specialized framework. Reliability comes from the complete system and its continuous evolution.
Safeguards
A chain of controls against hallucinations
A hallucination occurs when AI presents information as true even though it is not supported by the sources used. Palen layers several controls so reliability never depends on a single instruction or on the model alone.
A closed documentary environment
The AI does not search the Web. It works within a regulatory library reviewed, controlled and maintained by Palen for the supported countries and ADR editions.
Mandatory research before answering
The response path requires a search in that library. The AI must ground its answer in retrieved passages rather than bypassing the documentary step.
Dangerous goods safety adviser reflexes
The AI reasons like a dangerous goods safety adviser, not like a general research AI.
Evidence behind important claims
Regulatory statements carry a reference and Palen displays a consistent source list.
Permission to stop
When the evidence is insufficient, the AI must state the uncertainty instead of filling the gap. Empty, failed or technically incomplete responses are detected and removed.
Results monitored over time
How to read the 99.99% rate
This indicator measures the share of completed analyses with no hallucination detected by Palen’s internal monitoring or reported by a user.
Across 485 analyses completed since June 2026, no hallucination was detected by Palen’s internal monitoring or reported by users.
Why does Palen display 99.99%?
Palen deliberately keeps a margin of uncertainty. No AI system is infallible.
The indicator covers Palen’s internal monitoring and user reports since June 2026.
Safeguards reduce risk; they do not make AI infallible
These mechanisms target hallucinations. A sourced answer can still be unsuitable when the initial situation or the applicable regulatory context is incorrect.
- An incomplete description can lead to a sourced answer for the wrong situation.
- The ADR edition, corrections and national rules must correspond to the date and country concerned.
- A correct quotation can still be misapplied if a definition, exception or cross-reference is overlooked.
Test a sourced ADR analysis
Describe a real situation, review the assumptions and use the cited references to complete your professional verification.