2026-09-10 · Fuente：HotHardware
Jacob Coxon, who spent three years leading pretraining efforts, says unaligned superintelligence poses a genuine extinction threat to humanity within the next decade, reigniting the existential risk debate.
Jacob Coxon, an AI safety researcher who has resigned from Anthropic, is warning that humanity faces a genuine threat of extinction from unaligned superintelligence within the next decade — a horizon that places the projected catastrophe at roughly 2036. The resignation and its accompanying alarm have set off another round of argument over existential risk, turning a personnel departure into the latest flashpoint in a recurring industry fight.
Coxon is described as a key researcher who spent three years leading pretraining efforts — the large-scale training work that underpins frontier model development. That tenure places him among the practitioners closest to the systems whose trajectory he now questions, and his exit from one of the field's most prominent developers gives the warning an insider's framing rather than that of an outside critic.
The core of Coxon's claim is stark: unaligned superintelligence — systems exceeding human capability whose goals are not aligned with human interests — could pose a real extinction threat to humanity within ten years. Warnings of this kind are not new, but the report frames Coxon's intervention as notable precisely because it comes from someone who has just stepped away from hands-on work at a leading AI lab.
For readers tracking AI risk governance, the episode underscores how existential-risk arguments continue to surface from inside the industry's own labs, not only from external critics. Whether Coxon's 2036 timeline persuades skeptics or deepens existing fault lines, his departure adds another data point to a debate that shows no sign of resolution — and that now carries the signature of a former pretraining lead.
Etiquetas: Anthropic · Jacob Coxon · AI Safety · Existential Risk
萬安算交所 AIXX · AIXX · Exchange verificado · Ir al mercado