Proton cancelled its AI re-interviews, but the performance pressure remains

Proton has cancelled planned re-interviews in which nearly every employee was expected to explain how they use AI. According to 404 Media, the decision followed its investigation and internal resistance by one day. The retreat changes the review format, not the company's direction.

A two-hour company call ended the re-interview plan

404 Media based its report on conversations with 6 current employees and audio from an all-hands meeting lasting roughly two hours. CEO Andy Yen acknowledged a communication failure and announced that the re-interviews were cancelled. He also said he would continue pushing AI adoption.

Yen promised that a reorganisation planned for next month would cause no AI-related job losses. Employees nevertheless told the publication that measures intended for the re-interviews would become part of ordinary performance reviews. Proton did not respond to 404 Media's request for comment.

A company selling privacy is judged more harshly on speed

Proton built its brand around security and trust. A conflict between rapid AI deployment and careful testing therefore carries more weight than it would at an ordinary software startup. Customers are paying for the assumption that internal pressure will not overrule security judgement.

The re-interviews also turned AI from a tool into a personnel loyalty test. An employee could be judged by how much technology they used even when their role or security responsibility required caution. Moving the same pressure into performance reviews would change the form rather than the incentive.

Cancelling one meeting does not resolve the security conflict

The investigation relies on anonymous employees, leaked messages and internal audio. It contains serious testimony, but it is not an independent security audit of Proton's code. The available evidence does not establish a specific vulnerability in its products.

The management dispute is well documented. The report says security objections were mocked and leadership retreated only after publication. An apology for communication will carry little weight if incentives still punish people for slowing an AI rollout when they see risk.

Code review and personnel rules will show whether anything changed

A credible correction would appear in concrete safeguards: mandatory testing, clear responsibility for AI-written code, a safe route for escalating objections and employee evaluation based on outcomes rather than the number of AI tools used.

The second signal will be next month's reorganisation. If it avoids layoffs but preserves fear and the same metrics under another name, Proton will merely have removed the microphone from the stage. The conflict will continue in the offices.

Lilith's verdict

Proton removed the re-interview from the meeting room, but employees say its measuring stick remains with performance reviewers. A security company rebuilds trust by protecting the right to call stop in time, not by apologising for tone.