Proton cancelled its AI re-interviews—but moved the same test into performance reviews

Proton chief executive Andy Yen cancelled the planned near-company-wide re-interviews on AI use after 404 Media published internal messages and employee accounts describing fear, security objections and pressure to demonstrate enthusiasm. Six current employees and audio from a two-hour all-hands meeting confirmed the reversal. Yen acknowledged a communication failure and said he would reflect on how the programme had been presented. Chief technology officer Bart Butler told employees: “We fucked up.”

The policy changed less than the announcement suggests. Yen said he would continue pushing AI adoption, and employees were told that the criteria intended for the interviews would become part of ordinary performance reviews. One worker described the cancellation as damage control; another regarded it as real progress. Proton did not respond to 404 Media's follow-up questions. Users also reacted: some publicly cancelled subscriptions after the earlier investigation showed executives dismissing a request for several days of penetration testing as “loser thinking”.

Moving a test from a special interview into the standard review cycle may reduce the immediate spectacle and arbitrary scheduling. It can also make the pressure less visible. Performance systems are recurring, affect pay and employment, and often lack the evidentiary standard of a formal technical evaluation. If the metric remains whether an employee visibly uses AI rather than whether a workflow improves safely, the organisation preserves the incentive to optimise adoption theatre.

The useful correction would separate three questions: which tasks changed, what measurable benefit followed and which defects or security incidents the change produced. Those results should inform performance only where an employee controls them and where teams can report risk without retaliation. Proton's reversal proves that outside scrutiny can alter management behaviour quickly. It does not yet show that the company has replaced loyalty testing with outcome testing.