What happened
In an investigation published October 29, 2026, Joseph Cox of 404 Media reports on Proton’s internal push to increase AI use in engineering. Proton is the Swiss company behind Proton Mail and other privacy- and security-focused products, including encrypted storage, an authenticator, and a password manager. The reporting draws on interviews with eight current employees and one former employee, as well as Slack messages, meeting recordings, and an AI manifesto written by the company’s CTO.
The article recounts an exchange in a security Slack channel at the end of August. An employee urged the company to allow a few days to penetration-test newly written code before releasing it. CEO Andy Yen rejected the request, describing it as “loser thinking” and arguing that speed and security could coexist. He urged the team to stop handcrafting all code and use AI more. The investigation presents this exchange as one example of a broader executive effort to increase the company’s development velocity.
According to 404 Media, Yen created an email address for employees to report obstacles to AI adoption, with a stated aim of removing barriers to “AI velocity.” The company also installed software on workers’ computers to track how much AI-generated code it produces. The article says Yen told staff that 2026 would be Proton’s last year hiring engineers, after which it would hire people to build things with AI. It also reports that Proton had said it would re-interview staff about their AI use, with employees understanding that their willingness to adopt AI could affect whether they kept their jobs.
The investigation describes employees’ concerns about developing security products with AI and about speaking up. Internal material quoted by 404 Media says Proton was not on track to meet its annual recurring revenue target because of insufficient velocity, and characterizes the company’s growth as below market expectations. In response to detailed questions, a Proton spokesperson did not dispute the reporting or its characterization. The spokesperson said some employees were unhappy with the direction or misunderstood it, acknowledged anxiety, and said the company had sometimes been too blunt or too slow to explain itself while trying to move quickly.
What the archive adds
This is a new reported development in a thread already covered in Proton CEO cancels AI 're-interviews' of staff after 404 Media report on aggressive internal AI push. That earlier note described the cancellation of the re-interviews after scrutiny; this investigation provides a broader account of the internal AI push, including code-generation monitoring and reported security concerns. It adds detail to a known story rather than opening an unrelated one.
Trend signal
The case supports AI-native organizations move toward review and orchestration: increasing generated output shifts the organizational challenge toward specifying, routing, verifying, and owning that output. The investigation does not establish how Proton’s code performs, but the reported disagreement over security testing illustrates why velocity targets alone are an incomplete operating model.
Why now
The article connects the push to Proton’s stated concern that it would miss its annual recurring revenue target because of insufficient velocity, alongside growth that leadership described as underperforming. That context helps explain the urgency in the internal messages reported by 404 Media, though it does not prove that AI adoption will close the revenue gap.
Connections to other events
A separate earlier example in the archive described Meta discussing organizational change amid a 220% year-over-year increase in code changes after wider AI-tool use. Proton’s situation is distinct: the reporting centers on security objections and workplace pressure, not a measured increase in code output. Together, the accounts raise a practical question for engineering leaders: how to capture gains in production without reducing review capacity or making it harder for employees to report risk.
Why it matters
For teams building with AI, the relevant lesson is not that AI-generated code is inherently unsafe, nor that a particular testing window is always sufficient. It is that shipping faster changes the distribution of work: more code may arrive for review, while the people responsible for security still need time, authority, and psychological safety to challenge it.
Proton’s products make the tension unusually visible because users rely on them for privacy and security. A company can monitor adoption or set ambitious targets, but those metrics cannot substitute for evidence that changes have been tested and that failures can be raised before release. Teams adopting coding agents should define release gates that apply regardless of whether code was written by a person or generated with AI, and track review quality alongside output. The investigation is a reported account of one company, not proof that its approach is representative or that AI caused any specific security defect.
Limits and uncertainty
The source is an investigative report based on interviews and internal materials; the full Slack context, recordings, and manifesto are not reproduced in the supplied article excerpt. Proton did not dispute the reporting in its response, but that is not independent verification of every characterization or employee account. The material does not quantify Proton’s AI-generated code, establish defect rates, describe the software’s tracking methodology, or show whether AI use has improved delivery speed or revenue. It also does not establish what security testing was ultimately performed on the code in the cited exchange.
Reader takeaway
AI adoption becomes an engineering operating model when it changes incentives, measurement, and release decisions—not just the tools developers use. If a team asks for more generated code, it should preserve the capacity and authority to inspect that code, especially in security-sensitive systems. The Proton investigation is a warning about governance and trust, not a benchmark of AI coding performance.