The AI Act may not reach most workplace surveillance - The Next Web

The OECD has surveyed 6,047 firms across six countries and arrived at a number that ought to be read aloud in a committee room, ideally by someone holding a gavel: 67% of American firms use software to sanction underperforming staff, against 4% in the four European countries surveyed. That is not a gap, it's a canyon with a gift shop. The OECD, admirably deadpan, attributes the difference to "regulatory architecture" — which is Brussels-speak for "the Europeans wrote something down and the Americans wrote a bonus package."

The AI Act, that majestic doorstop of a regulation, was drafted to civilise artificial intelligence. Excellent news, then, that it turns out to be a nightclub with a very strict guest list. The OECD's own caveat is that most of these sanctioning tools aren't AI at all, which means the machinery quietly deciding who gets a stern email and who gets shown the door sits just outside the velvet rope, entirely unbothered. You were not managed by an algorithm after all. You were managed by a spreadsheet, a dashboard, and a manager who has outsourced the unpleasant bit. The algorithm was merely the vibe.

The genuinely impressive part is the efficiency of the blind spot. Nobody had to lobby for it. Nobody had to water anything down. The law simply defined its subject narrowly enough — AI, and only AI, and only in the ways we enumerated — that a great deal of workplace surveillance now happens in the linguistic gap between what the technology is called and what it does. Six thousand and forty-seven firms, six countries, one lesson: legislate against a noun, and the industry will cheerfully change the noun. Next quarter's procurement deck will say "analytics." The workers will still be sanctioned. The architecture, as ever, holds.