Principles for the use of AI
Page reviewed on
In brief
Nine commitments applied by AGGIL to the use of artificial intelligence in its assignments: written client consent, services that do not train on the data, human review and signature, verification against the source, logging. AGGIL issues no certification.
No client document is submitted to an AI service without the client's written consent.
How to check: The consent is kept in the assignment file.
Only services whose terms exclude training on the data submitted are used.
How to check: The services used are named in the deliverable (commitment 6).
Documents from a tax audit are never submitted to a hosted AI service.
How to check: The assignment log allows this to be checked (commitment 7).
Every deliverable is reviewed and signed by Fabrice Trollet, who is accountable for it; an AI is never its author.
How to check: Signature on the deliverable.
No reference, quotation or figure enters a deliverable without verification against the primary source.
How to check: Each reference in the deliverable states its source.
The deliverable tells the client which tasks were given to an AI and which models were used (name, version, date).
How to check: Dedicated section of the deliverable.
When an AI analyses documents, the work is logged and kept for at least six months, available to the client.
How to check: The log is provided to the client on request during that period.
AGGIL does not certify the compliance of an AI system and issues no certification.
How to check: No deliverable bears a certification statement.
The use of AI is disclosed in publications.
How to check: Statement in each publication concerned.
Compliance and reliability of agentic AI systems · Website privacy policy