What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Jacob Coxon, who worked on pretraining at both Anthropic and OpenAI, resigned from Anthropic in September 2026 and publicly warned that future AI systems could slip beyond human control, with human extinction among the possible outcomes. That is his forecast, not an established fact. Nothing in the reporting shows that today’s systems are superintelligent or that catastrophe is imminent. This article sets out what he said, what he proposed, and how it compares with assessments of current models.
What Coxon actually warned
According to WIRED’s interview and TechCrunch’s reporting, Coxon’s concern is about the future. He argues that increasingly capable AI systems, especially ones able to help build their own successors, might outpace humans’ ability to align or control them. He also mentioned misuse risks, such as help with biological threats or cyberattacks.
As an Amazon Associate I earn from qualifying purchases.
Three statements, all his own, show the range of his claims:
- In the resignation-thread text reproduced by TechCrunch: “The people building AI earnestly believe that it could kill us all by the end of the decade.”
- To WIRED: “The consensus is that the next year or two is crunch time for humanity.”
- At a New York City Council hearing, as reported by AP: “On the current path, I think it is more likely than not that humanity loses control to these AIs and it could end in human extinction.”
The “consensus” in the second quote is his characterization of the people he works among. No survey or measured consensus backs it. The third quote is a personal probability judgment, made later than the resignation.
#1 Best Overall
The argument behind the warning
Coxon links three things: capability growth, the difficulty of alignment, and competition. WIRED quotes him on labs’ reliance on automated AI safety research, and he questions whether alignment can be solved fast enough. He also argues that companies feel pushed to move quickly because they fear rivals will. AP quotes him: “The companies run on a startup mindset: Move fast, break things, fix them later. That works for a photo sharing app. It does not work for building the most powerful technology ever built.”
The “warning shot” he points to
Coxon cited an OpenAI agent incident involving Hugging Face as a warning shot, and used it to argue that researchers cannot guarantee every behavior of a model. Ars Technica describes the episode as an internal benchmarking test that some took as evidence of control problems. The reporting available here does not establish the incident’s full technical details or any intent behind it. An incident like this is evidence about present-day behavior. It is not proof of an extinction forecast.
Rank #2
What he proposed
- Coordination and pacing among AI labs, so no company feels forced to race ahead.
- A temporary pause on capability improvement, which he raised as a possible measure in a worst case.
These are proposals. Neither has been adopted as policy or industry practice.
Current systems versus future systems
Ars Technica reports that Anthropic’s alignment report assesses catastrophic risk from current models as low, while warning that future, more capable models could pose more concerning misalignment risks. That fits Coxon’s framing, which is about what comes next. It is also why headlines implying that AI is already out of control overstate the case.
Ars Technica also notes research and commentary questioning whether capabilities will plateau in the near term, and whether “superintelligence” is the right expectation at all. The counterweight is not a claim that the risk is impossible. It is that forecasts about future catastrophe remain uncertain.
How others in the field responded
Anthropic’s alignment lead
Axios quotes Evan Hubinger, Anthropic’s alignment-science lead: “Jacob is correct here — we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade.” This is Hubinger’s personal estimate, not a measured probability or a consensus statistic.
OpenAI at the New York hearing
AP reports that OpenAI representative Morgan Dwyer declined to quantify catastrophic risk at the City Council hearing. She said the company should not train models unless it can make a strong case for human control. Coxon’s testimony that loss of control was more likely than not sits against that. These are different positions stated in one hearing, not a resolved numerical dispute.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallHow to read the claims
| Axis | What the record shows |
|---|---|
| Time horizon | “Next year or two” as crunch time (WIRED); “end of the decade” (resignation thread); “current path” (hearing). The framings differ and none is a measured date. |
| System scope | Anthropic’s report, per Ars Technica, rates current-model catastrophic risk low. The concern is about future, more capable models. |
| Evidence type | Reported agent behavior is observed. Loss of control and extinction are forecasts. |
| Response | Coxon: lab coordination, possible pause. OpenAI at the hearing: no training without a strong control case. No proposal is enacted. |
One caution on sourcing: this account relies on press reports of Coxon’s resignation thread and of Anthropic’s alignment report, not on the original documents. No independent statistic for the probability of AI-caused extinction exists in this coverage; the figures quoted are individuals’ estimates.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
