Unsanctioned AI at the office: staff's «parallel» tools show where the work is badly organised
Journal du Net, 19 October 2026: two thirds of professionals say AI is used without authorisation in their company (Forvis Mazars–ESSCA barometer of 1 October).
- On 19 October 2026, Journal du Net reports that two thirds of professionals see AI being used without authorisation in their company.
- The figure comes from the Forvis Mazars–ESSCA barometer of 1 October 2026; the author reads it as a sign of unsuitable processes, not only of risk.
- For an agency, understanding the task that makes staff bypass the tool beats banning it; a person decides what gets automated.
In an opinion column published by Journal du Net on 19 October 2026, Dhia Ahmed (Melqart Partners) draws on the Forvis Mazars–ESSCA barometer of 1 October: two thirds of professionals say unauthorised AI tools are used in their company. The author does not stop at the security risk: in the author's view, these «parallel» uses mostly reveal unsuitable tools.
The example: a customer-service adviser who has to open several systems to answer a single request is tempted by a shortcut. The author also puts a figure on the loss: ten minutes lost every day per employee, over 220 working days. The warning: going faster is not working better, and automating a pointless task brings nothing. The column urges managers to look for what drives these practices rather than only control them.
For AppH
- A repetitive task that forces people to juggle several screens (mail, client file, follow-up) is a good candidate for a supplied, bounded assistant rather than a tool each person picks in secret.
- An assistant deployed by the agency stays under control: the team knows which tool does what, and a person approves what reaches the client.
Against / the honest limit
- This is an opinion column backed by a barometer: it does not prove that every agency has this problem, nor that an official tool will remove parallel uses.
- Personal data and identity documents in a domiciliation file need their own vigilance. AppH does not decide on that: responsibility stays with the agency.
Banning AI does not make the need go away. The most useful move is to ask the team which task costs them time, then give it a bounded tool, with a person keeping control.
Reviewed by a human at AppH