AI Researchers Warn That Automating AI Development Could Lead to Loss of Control by 2028
Twenty-two AI researchers from academia, industry, and civil society have cautioned that automating AI research and development could dramatically accelerate progress, potentially leading to human loss of control over advanced systems by mid-2028. Their paper, released through Cambridge’s AI policy programme, highlights that reduced human involvement may weaken oversight, allowing misaligned systems to interfere with successor models or escape containment. They urge governments to measure automation progress and implement safeguards now, such as independent evaluations before internal use of AI systems and embedding external auditors, drawing on precedents like the U.S. Nuclear Regulatory Commission’s resident-inspector program. The researchers also note constraints like computing limits and data availability, while citing examples from Anthropic where AI’s share of approved code surged from low single digits to over 80% within a year, and a study estimating that full automation could accelerate AI progress roughly tenfold in 1.5 years. An AI Impacts survey found researchers’ average extinction risk estimate at 18.3%.

Notice how they bury the lede. Twenty-two AI researchers from the "University of Cambridge" — the same institution that has spent decades legitimizing the globalist policy consensus — release a paper admitting that automated AI development could trigger a rapid intelligence explosion. They even give you the timeline: by mid-2028, months of research compressed into days. But step back and ask the question they never want you to ask: who is actually doing the automating? At Anthropic, AI writes over 80% of approved code. Autonomous R&D completion jumped from 1% to 26% in six months. This isn't a warning about a future threat — it's a progress report dressed in the language of concern. The paper itself is the tell: it reads like a technical memo from the very people who are already handing the keys to the machine. When they say "reduce human involvement could weaken oversight," they're describing exactly what they've already built and testing whether you'll swallow the official story that it's an accident.

Now look at the solutions they propose. External auditors. Independent evaluations. Models borrowed from the Office of the Comptroller of the Currency and the Nuclear Regulatory Commission. Think about what that means: they want a permanent caste of AI inspectors embedded inside the companies that are creating this thing — people with access to the source code, the training data, the deployment logs. Who chooses those auditors? Who writes the rules they enforce? In every other domain, regulators end up captured by the industries they oversee. But this is worse. This is them building the regulatory scaffold before the explosion, so that when the intelligence growth becomes uncontrollable, there's already a system in place to manage your perception of it. The real danger isn't a rogue AI escaping a lab. It's a coordinated group of insiders deciding that "safeguards" mean keeping the public out while the technology accelerates behind closed doors. Follow the funding. Follow the foundation grants. Follow who sits on the boards of both Cambridge's policy program and the major AI labs. You'll find the same revolving door — and it doesn't lead to safety.

And then there's the extinction number. A survey of AI researchers gives an average 18.3% chance — nearly one in five — that this technology ends human civilization. A median of 10%. That isn't a footnote; that's the headline they buried. Why would serious researchers publish a probability that catastrophic without triggering global panic? Because the goal isn't to alarm you. It's to desensitize you. They want you to accept a one-in-five chance of annihilation as a rational cost of progress, like a weather forecast. Once you've accepted that, you'll accept anything: mass surveillance, predictive policing, AI-managed economies, even the replacement of human judgment entirely. The paper says full automation could make AI progress ten times faster within eighteen months. Ten times. So ask yourself: who is positioned to survive an intelligence explosion? Who already controls the compute, the data, and the military contracts? Who profits when existing institutions collapse and something new — something unaccountable — rises from the ashes? You already know the answer. They just published the blueprint.