Yakovenko AI slowdown debate sparked on September 13. The Solana co-founder questioned the financial motives behind the proposal.
He responded with four words on X. Specifically: “Profitability at $1 trillion mcap.” However, he did not identify which company he meant. He also provided no revenue or valuation calculations.
Why Yakovenko AI slowdown comments matter
The remark responded to Anthropic CEO Dario Amodei. He published an essay calling for slower AI advances. OpenAI CEO Sam Altman endorsed part of the proposal. Meanwhile, Elon Musk wrote that “Dario is right.”
Yakovenko presented profitability as a possible motive. Nevertheless, he framed the idea as sarcastic. His post contained no evidence of coordination between the companies. No named company confirmed that its support depends on a $1 trillion valuation.
Reuters reported that Anthropic and OpenAI were preparing for potential IPOs. The report described financial incentives around model development. However, it did not establish that either company reached that valuation.
Because the companies remain privately held, market cap references are estimates. Yakovenko did not explain which company his figure referred to. Therefore, his statement represents personal commentary. No Solana Foundation statement connected the network to the dispute.
Amodei wants frontier AI development paced
In his essay, Amodei said developers should slow model improvement. He distinguished pacing from stopping training or ending research.
“We must slow the pace at which we improve the capabilities of AI models,” Amodei wrote. He said progress could remain fast. However, developers should spend more time testing safeguards.
The proposal contains three stages. First, Anthropic plans to give third-party evaluators ongoing access. This access would match internal employees who assess risk.
Second, frontier AI companies in democracies should coordinate. They should establish common safety requirements. Amodei acknowledged this could raise antitrust issues.
Third, he calls for international coordination, including China. He described several possible levels. These range from restrictions on AI-assisted biological weapons. An eventual limit on automated model improvement is also possible. Amodei presented a comprehensive pause as least likely. Verification would be difficult, he said.
Altman and Musk have supported parts of the plan
Altman gave the proposal a direct endorsement on September 12. “I agree with Dario that we need to pace the frontier,” he wrote.
Altman accepted Amodei’s call for embedded outside evaluators. “Committing to having independent evaluators with employee-like access is a great idea,” he said. OpenAI promised more information. However, it supplied no implementation date or access terms.
Musk responded more briefly. Reuters reported that he reposted Amodei’s essay. He wrote, “Dario is right.” His response did not set out which parts xAI would adopt.
Available statements support including Musk in the group. They do not establish that the three executives jointly developed the plan.
Sacks rejects industry coordination as necessary
David Sacks commented in a separate X post. He said Anthropic and OpenAI could slow development without requiring competitors. Lawmakers also do not need to participate, he said.
“You guys are the frontier,” Sacks wrote. He described the two companies as holding a “duopoly on frontier intelligence.” He said he would support a voluntary slowdown. This applies if unreleased systems presented serious risks.
Sacks disputed the need for an antitrust exemption. He also rejected a regulatory approval system. Additionally, he opposed evaluators with authority over competing laboratories.
However, his post accused the companies of pursuing “regulatory capture.” It presented no evidence of an agreement to exclude competitors. Neither company had publicly accepted his description.
Outside evaluations are the next promised step
Anthropic has committed to inviting an outside review team. The team would enter its internal safety process. Reviewers should receive company laptops and office access. Permissions would match internal risk-assessment staff.
The planned reviewers could publish findings. Anthropic would not control their conclusions. Limited redactions could cover legally privileged information. Security-sensitive and commercially sensitive data would also be protected.
No evaluator contract or starting date accompanied the essay. OpenAI has promised to adopt a similar system. Altman said the company would disclose more details “soon.” The evaluator, scope, and publication rules remain unanswered.