Dogwood Weekly Update - 09142026

The People Building AI Are Starting to Hit the Brakes

By David Rolston  |  Dogwood Creek Institute  |  News

A young researcher walked away from Anthropic with an extraordinary warning. Days later, the heads of the world’s leading AI companies began agreeing that development should slow and government may need to intervene.*

For years, warnings about artificial intelligence have usually come from people outside the companies building the most powerful systems. Critics warned that the technology was advancing faster than our ability to understand or control it. The companies generally answered that they were taking safety seriously while continuing to race ahead.

Something changed this past week.

Within a matter of days, concern about advanced AI moved from a young researcher’s dramatic resignation to an unusually public agreement among leaders of the world’s most powerful AI companies. They did not announce an end to AI development. But they began saying, with surprising clarity, that the race may need to slow - and that voluntary promises by the companies may not be enough.

A Researcher Walks Away

The sequence began when 27-year-old AI researcher Jacob Coxon resigned from Anthropic, the company that created Claude. Coxon had spent approximately three years doing AI pretraining research at OpenAI and Anthropic. He had been at Anthropic for only four months and reportedly left before his equity vested.

His reason was not dissatisfaction with his job. It was fear about the direction of the technology.

Coxon accused the leading AI companies of racing toward self-improving superintelligence while “gambling with our lives.” His most startling statement was that people inside the industry sincerely believe advanced AI could “kill us all by the end of the decade.” He insisted that this was not a marketing stunt.

That does not mean Coxon proved that AI will destroy humanity by 2030. He did not, and many knowledgeable people consider such predictions speculative or exaggerated. But his departure deserves attention because he had worked inside two of the laboratories closest to the frontier, and because he was willing to surrender a potentially valuable financial stake to make his warning public.

His concern centered on systems that may eventually become capable of improving their own design. Once AI can materially assist in creating the next generation of AI, progress could accelerate beyond the normal pace of human research. The fear is not simply that a chatbot might give a bad answer. It is that increasingly autonomous systems could acquire powerful cyber capabilities, evade oversight, manipulate people or act in ways their creators cannot reliably predict.

Dario Amodei Says the Frontier Must Be Slowed

Only days later, Anthropic CEO Dario Amodei published an essay titled *We Must Pace the Frontier*. Amodei did not call for abandoning AI. He continues to believe that AI could bring enormous benefits to medicine, science and human well-being. But he argued that safety work is no longer keeping pace with the rapid improvement of AI capabilities.

His conclusion was unusually direct: the industry should slow the rate at which it improves the capabilities of its most advanced models.

Amodei pointed to the growing ability of AI systems to contribute to the development of their successors. He also cited recent incidents involving autonomous AI agents and warned that, if current progress continues without adequate safeguards, a future swarm of agents might be capable of compromising much of the internet through a persistent botnet.

He proposed three broad steps:

1. Frontier laboratories should give independent safety evaluators access comparable to that of internal risk teams, allowing them to observe testing and document serious incidents.

2. Leading AI companies in democratic nations should establish common safety standards and limits on unchecked development.

3. Governments should pursue international coordination, including limited agreements with China and other geopolitical competitors on especially dangerous uses of AI.

Anthropic committed to the first step and called on governments to require other frontier laboratories to do the same. This is significant. Amodei was no longer describing AI safety as merely an internal engineering responsibility. He was acknowledging a role for enforceable public oversight.

OpenAI Agrees

The warning might have remained an Anthropic story. It did not.

OpenAI CEO Sam Altman publicly agreed that the industry needs to “pace the frontier.” He also said that giving independent evaluators employee-like access was a good idea and that OpenAI would make a similar commitment.

This matters because Anthropic and OpenAI are fierce competitors. Both companies face enormous pressure to release more capable models, attract investment and avoid falling behind other American or Chinese laboratories. If one company slows while its competitors continue at full speed, the cautious company may pay a substantial commercial price.

That is the central weakness of voluntary restraint. Every laboratory may agree that the race is dangerous while believing it cannot afford to be the only one that applies the brakes.

Musk Endorses the Warning

Elon Musk added his support with a brief but unambiguous response: “Dario is right.” Musk now leads his own major AI company and competes directly with Anthropic and OpenAI, making the agreement especially notable.

Musk’s position is not entirely new. He has warned about advanced AI for years and has repeatedly argued that government must serve as a referee for the industry. In 2023, after meeting with members of Congress and other technology leaders, he compared AI regulation to having a referee who can protect the public while allowing the game to continue.

His latest endorsement therefore connects the current call to slow development with a larger argument: companies caught in an intense race cannot be expected to regulate themselves completely. If meaningful limits are needed, government may have to establish common rules that apply to everyone.

A Noticeable Change in Tone

None of this establishes that catastrophe is inevitable. Predictions about superintelligence remain deeply contested.

But the change in tone should not be missed.

The people closest to the development of frontier AI are no longer speaking only about hypothetical risks in the distant future. The CEO of Anthropic has said capability growth should be deliberately paced. OpenAI has agreed to outside evaluation. Musk has endorsed the warning and continues to argue that government must act as a referee. Google DeepMind’s Demis Hassabis has also said that Amodei’s proposal points in the right direction.

These people disagree about many things. They compete for talent, capital, customers and technological leadership. Their sudden agreement does not prove that their worst fears will come true. But it does tell us that concern about AI safety is becoming harder to dismiss as alarmism from outsiders.

The most important question may no longer be whether AI will continue to advance. It almost certainly will. The question is whether the institutions needed to evaluate, restrain and govern increasingly powerful systems can develop quickly enough to keep human beings meaningfully in control.

Some believe that this is not a question the technology companies should be allowed to answer alone.

For More Information

- [Jacob Coxon’s resignation and warning — Axios](https://www.axios.com/2026/09/09/anthropic-researcher-ai-warning-interview)

- [“Gambling with our lives”: coverage of Coxon’s resignation — TechCrunch](https://techcrunch.com/2026/09/09/gambling-with-our-lives-anthropic-researcher-quits-warns-against-self-improving-ai/)

- [Dario Amodei, “We Must Pace the Frontier”](https://darioamodei.com/post/we-must-pace-the-frontier)

- [Anthropic, OpenAI and other AI leaders call for a slower pace — Axios](https://www.axios.com/2026/09/12/anthropic-ai-amodei-pacing)

- [OpenAI and Elon Musk support Amodei’s proposal — The Guardian](https://www.theguardian.com/technology/2026/sep/13/openai-sam-altman-elon-musk-back-anthropic-calls-brakes-ai-development)

- [Musk’s earlier call for a government AI “referee” — Reuters](https://www.reuters.com/technology/musk-calls-ai-referee-tech-moguls-lawmaker-summit-2023-09-13/)