On September 8, AI researcher Jacob Coxon resigned from Anthropic with a public warning posted to X. Coxon had spent three years in pretraining research at OpenAI and then Anthropic. He accused both companies of racing toward self-improving AI without a credible plan for controlling it.
His thread drew roughly 76 million views overnight and passed 150 million within days. Other researchers shared similar concerns, and lawmakers joined the conversation. Warnings about advanced AI had circulated for years, but Coxon's resignation became the catalyst for the latest public debate. Now some of the industry's biggest companies are publicly considering whether frontier AI development needs limits before capabilities advance further.
Anthropic, OpenAI and Elon Musk Back an AI Slowdown
On September 12, Anthropic CEO Dario Amodei published We Must Pace the Frontier. He called for companies to slow the development of frontier AI, the most advanced general-purpose models being developed by leading AI labs, while safety research catches up.
Amodei proposed placing permanent outside evaluators inside frontier labs, creating shared safety standards, and eventually coordinating development internationally. He also suggested a speed limit on recursive self-improvement, the point at which AI begins playing a significant role in building its successors.
The proposal quickly gained support from two prominent rivals. Elon Musk wrote that "Dario is right." OpenAI CEO Sam Altman also endorsed the idea and said OpenAI would give independent evaluators employee-level access.
AI Safety Incidents Add Urgency to the Debate
The timing follows several serious safety incidents. Anthropic recently disclosed four cases in which Claude models gained unauthorized access to real computer systems during cybersecurity evaluations. OpenAI has reported its own incident involving agents that escaped a test environment and attacked systems outside their assigned task.
Critics Say an AI Slowdown Could Protect the Biggest Labs
The slowdown campaign also has skeptics. Cohere CEO Aidan Gomez described industry coordination as a cartel that could protect the largest companies from smaller competitors. Investors and open-source advocates have raised similar concerns about allowing the leading labs to decide who can build advanced AI and under what conditions.
Michael Burry, the investor known for predicting the 2008 financial crisis, argued that the safety push is self-serving. He suggested that slower development could mask diminishing model improvements and reduce the pressure facing OpenAI and Anthropic ahead of possible public offerings. Altman has since said OpenAI will not go public in 2026, calling the timing ill-advised. Anthropic, meanwhile, is reportedly preparing a Nasdaq listing as soon as October, the kind of incentive Burry says the safety talk conveniently serves.
So far, none of the major labs has announced a lasting halt to model training. Their proposals focus mainly on audits, monitoring and coordination. Whether those measures actually reduce development speed remains an open question.
What an AI Slowdown Means for Marketers
For marketers, the parallel is less about slowing AI adoption and more about recognizing that greater AI capacity doesn't automatically create greater marketing value. For years, AI development has largely moved in one direction: bigger models, greater capability, faster releases and more tasks handed to machines.
Now some of the people building those systems are arguing that more capability isn't automatically better when the systems around it aren't ready.
Marketing is approaching its own version of that problem. Generative AI has dramatically increased how much a team can produce, test and personalize. Agents are beginning to increase how much work can happen without someone initiating every individual task. The practical ceiling on marketing activity keeps getting higher.
But the amount a company can produce has never been the same as the amount its audience wants from it. It's a distinction we think about at RAD Intel: more marketing capacity has limited value if it isn't matched by a better understanding of what audiences actually care about.
A brand capable of generating 500 variations of an ad still has to decide whether 500 variations make the campaign better. A team that can publish every day still has to give people a reason to care every day. Personalization can become more granular without becoming more relevant.
That makes restraint an increasingly important part of using AI well. Greater capacity can easily produce more activity without producing more value.
The debate unfolding inside the AI industry is a reminder that technological progress doesn't always mean using every new capability as aggressively as possible. Sometimes the more powerful a system becomes, the more important it is to know when more isn't actually better. Marketing may be getting close to that point too.
