Who supervises workplace AI? Poland's implementation of the AI Act and the exit of the Labour Inspectorate

Regulation (EU) 2026/1744 defers the AI Act's obligations for high-risk employment systems to December 2027, a year after Member States must transpose the Platform Work Directive. The order in which EU rules on algorithmic management reach the workplace has been inverted, and the first national implementing measures are being adopted in the interval. Poland's Act of 3 July 2026 is one of them. This development note examines what it does with the obligation in Article 26(7) AIA, which requires employers to inform workers’ representatives before putting a high-risk system into service and refers the manner of doing so back to national law. Drawing on the two rounds of consultation on the Polish bill, it shows that the Ministry refused trade union representation on the new supervisory Commission on the ground that the Labour Inspectorate sat on it as a permanent member, and that the inspectorate was removed four months later, at its own request, invoking ILO Convention No 81. The refusal stood; the reason for it did not. The episode illustrates a problem that the Union instruments do not address: representation that is present on paper, relied upon to answer objections and then withdrawn.