Skip to content
Human by choice
A Policy Window advocacy programmeWiki · neutral reference ↗

Human by choice · Commentary · A Policy Window advocacy programme

Reading AI agreements for decision power

Prepared Published Edition 5

Before describing an AI proposal as giving workers control, identify the decision they can alter. Being informed, agreeing to a change and deciding where its benefits go are different roles.

One agreement contains different arrangements

BNP Paribas and UNI Global Union signed a group agreement on 4 November 2024. Its technology provisions commit to human supervision of AI, referencing a 2020 European framework. A separate reorganisation provision invites local management to inform representatives and involve employees upstream, subject to local rules. [1]

The teleworking section specifies voluntary participation by both employee and manager and reversibility, within employer compatibility assessments and local rules. This concerns teleworking; it does not demonstrate an AI opt-out or authority over savings. The agreement also states that it confers no contractual rights on employees or third parties. That does not settle rights under local law or other agreements. [1]

It provides local-first dispute handling, party-level escalation and mediation subject to the agreement's conditions. [1]

Read separately, the 2020 framework describes affected workers requesting human intervention or contesting a decision when AI is used in human-resources procedures. It also includes testing AI outcomes in that setting. It also calls for strategies that benefit both enterprises and workers, including sharing productivity gains. It provides for implementation through national, sectoral or enterprise arrangements. This does not establish an automatic share, a current remedy or incorporation of every provision into BNP Paribas’s agreement. [2]

Match the power to the decision

A role in checking an AI output, a choice about where to work and a say over how gains are shared each need their own terms. Look for the actor, the decision, whether agreement is required before a change, and what happens after refusal. Do not turn a provision about one decision into a promise about another.

For a new proposal, use the allocation statement to record actual powers and missing ones. The fictional worked example shows competing requests without inventing consent. These tools help readers ask what would have to change before they could say workers decide where AI's benefits go.

Sources

[1] French agreement, Chapters 3, 4 and 9, pp. 15–18, 31–34. French prevails; these passages were checked against English. This describes the 2024 text, not current legal entitlements or implementation.

[2] European Social Partners Framework Agreement on Digitalisation, June 2020, pp. 9, 11–13. The signed text was checked separately; its implementation and current application have not been established here.


Human by choice is Policy Window's advocacy programme on AI, work and who decides. This article is analysis and advocacy, distinct from the neutral Policy Window wiki.

Explore the Human by choice readings