A new report is getting attention for an unusual reason: the people signing it are not outside critics. They include Jakub Pachocki, the chief scientist at OpenAI, and Jack Clark, a co-founder of Anthropic, standing alongside AI pioneers Geoffrey Hinton and Yoshua Bengio. These are people whose companies profit from AI moving as fast as possible, and they are asking governments to slow down and watch closely.
The idea they are warning about is not new. A mathematician named Irving Good described it back in 1965: a machine smart enough to design a smarter version of itself could trigger a chain reaction, with each new version building the next one faster than humans could follow. For sixty years this stayed a thought experiment, and the report argues it is starting to look like a near-term possibility rather than science fiction.
Here is the part that should catch a business owner's attention. Anthropic has said AI now writes the large majority of its own code. The report's authors say some research tasks that currently take human teams months to finish could be fully handled by AI within a couple of years.
None of this means an out of control AI takeover is imminent. The report is careful to say the effect has not yet crossed the threshold needed to set off a runaway spiral, and real world bottlenecks like supply chains, lab equipment, and regulations would slow down even a very fast moving AI. But the authors argue that once the spiral starts, there may not be time to write the rulebook afterward.
Their proposed rules are practical, not dramatic. They want AI companies to publish regular reports on how much of their own research is automated, and to allow independent inspectors inside these companies, similar to how bank regulators work.
They also want the ability to pause a specific project if it moves too fast, plus emergency plans ready in advance, the same way governments plan for pandemics or natural disasters rather than writing the plan after the event. Notably, OpenAI and Anthropic have separately been discussing an agreement to test each other's AI models, a sign that even fierce competitors see the risk as real.
For a business leader with no stake in the AI industry itself, the practical takeaway is about timing, not technology. If research and development inside AI companies keeps compressing from years into months, the products, prices, and competitive advantages built on today's AI could shift on a similar timeline. That means treating your AI vendor relationships and cybersecurity plans as things to revisit every quarter, not once a year, rather than assuming this year's capabilities are a stable baseline.