2026-09-11 · Fuente：WinFuture
Two ex-safety researchers from Anthropic and Google DeepMind say advanced AI could slip out of human control — and that no one today has the ability to intervene.
Two former AI safety researchers — one previously at Anthropic, the other at Google DeepMind — have gone public with a stark warning about the trajectory of artificial intelligence. Their message, as reported by WinFuture, is blunt: advanced AI could slip beyond humanity's grasp, and at present nobody is in a position to regain control or intervene.
The core of the warning is loss of control. The researchers argue the technology could escape human oversight as it advances, and the report's headline sharpens the point further: nobody currently has the situation under control, and nobody would be able to step in. Framed that way, the risk is less a distant hypothetical than a present-day governance gap — a statement that reliable means of steering or restraining frontier systems, should they begin to behave in unintended ways, do not yet exist.
The report's wording notes that the pair have 'also' come forward — a small word with significance. It places them alongside other former safety staff from leading AI labs who have carried their concerns into public view rather than confining them to internal channels. Because Anthropic and Google DeepMind both develop frontier-scale AI, warnings from their alumni land with particular force: these are voices from inside the organizations building the most capable systems, not outside commentators.
For the AI compute sector, the intervention arrives amid an intense capital cycle. Frontier capability races are pulling vast investment into GPUs, accelerators and data-center capacity, and safety warnings voiced from inside the lab ecosystem feed directly into the regulatory and investor climate that funds that buildout. Alerts of this kind rarely halt procurement on their own, but they shape the policy environment — oversight regimes, liability questions, potential testing mandates — within which compute buyers and investors plan multi-year commitments.
The takeaway for readers tracking the buildout is straightforward: some of the people closest to frontier development are saying publicly that control, not capability, is the question nobody has yet answered.
Etiquetas: Anthropic · Google DeepMind · AI Safety · AI Risk · Frontier AI
萬安算交所 AIXX · AIXX · Exchange verificado · Ir al mercado