Gates revives call for an AI tax, but regulators still aren't listening

The useful thing about a memo is that it hands an official a sentence with a name on it. Hearings and vendor questionnaires run on attributable claims, and the line about the industry crossing safety lines it set for itself ^[2] is now attributable, in writing, to someone the industry cannot file under outsider. His GeekWire remark that the loss of control he expected to be many years off is already visible ^[4] does the same job in fewer words. Neither needs a legislature to have an effect.

Separate what is being pitched from what is being done. Gates is pitching fiscal and occupational limits: money taken off the top, and work fenced off for people ^[1]. Europe's actual instrument classifies risk and requires a human in the loop ^[10]. By TNW's reading, the AI Act taxes no model, reserves no occupation, and no member state has proposed either ^[11]. So the near-term effect of the memo is not adoption. It is raising the price of saying "we take safety seriously" inside a process that already exists, because the company answering an oversight question now answers it alongside a claim that the lines were self-set and got crossed.

The 2017 vote shows how the two halves come apart. Add the tallies and 604 MEPs recorded a position; the 396 in favour were 76.3 percent of the 519 who voted yes or no ^[13][6]. That is a comfortable majority for a text asking the Commission to examine whether member states' social security systems hold up as robots spread ^[7], with the tax already out of it, agreement on the diagnosis stopping well short of agreement on the remedy. The same resolution floated electronic personhood for the most sophisticated autonomous robots, and that has gone nowhere ^[8], which is a fair guide to what a striking paragraph in an adopted text is worth. The rapporteur, Mady Delvaux, said coalitions in the Parliament had rejected a forward-looking debate about workers, and the robotics industry welcomed the result ^[9]. Gates first put the robot tax up in February 2017 ^[5]; TNW's summary is that nine years on, the position has not moved ^[12].

The evidence itself gets less scrutiny from either camp than the diagnosis does. TNW argues the base is thinner than either side admits, and that what AI is doing to jobs in Europe looks unlike the headline numbers ^[14]. An operator cannot size their own exposure from a national forecast in any case. What is measurable inside one company is narrower: how often the human in the loop actually changes the output, how long the automated task takes now against what it took before, and whether the team that adopted the workflow is still using it a quarter later.

The practical test for a product team is a short inventory: every safety line the product claims, checked against who set it, you or a regulator, and against whether it could be shown to a stranger. The lines that turn out to be self-set and unshowable are the ones now exposed, since that is the category Gates just made quotable, and the one left explaining it on Friday is whoever shipped it.