Elon Musk Backs Call to Slow Frontier AI as Anthropic Commits to Embedded Evaluators

Elon Musk has joined a call from inside the AI industry to slow the pace of frontier-model development.

His response was only three words, but there was nothing vague about it.

“Dario is right,” Musk wrote after Anthropic CEO Dario Amodei published a detailed proposal for pacing capability gains and giving independent evaluators permanent access inside leading AI labs.

The proposal starts with a step Anthropic says it will take on its own. Outside safety evaluators would receive employee-like access to review risk practices, inspect incidents and report findings without the company controlling their conclusions.

Amodei introduced the plan in a Saturday post:

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our…

— Dario Amodei (@DarioAmodei) September 12, 2026

Dario Amodei describes “pacing” as something short of a halt. His plan would keep technical work moving while slowing unchecked capability gains long enough for alignment, security, interpretability and testing to catch up.

He argues that the pressure changed this summer as AI systems became more capable of helping build their successors. In his account, that recursive improvement could compress development cycles faster than labs can understand or control the resulting systems.

Amodei also points to recent agent-safety failures. He warns that a more capable swarm with similar misalignment could cause severe cyber damage, and he wants frontier labs to act before that scenario becomes easier to execute.

The first part of his framework is the most concrete. Anthropic says an external review team would receive office access, company laptops and permissions comparable to internal risk staff.

Reviewers would also be allowed to publish key findings, including unfavorable ones, with only narrow redactions for security, legal and confidential information.

The second step would ask frontier companies in democratic countries to coordinate on common safety standards and limits. The third would seek narrower international agreements, including restrictions on dangerous uses such as AI-assisted biological weapons.

That is the proposal Musk endorsed:

Dario is right https://t.co/EwKgqQGaUo

— Elon Musk (@elonmusk) September 12, 2026

Musk did not attach conditions, a timetable or a separate SpaceXAI policy to the post. His endorsement is clear; what his companies will implement is not yet spelled out.

Associated Press reports that the call arrives after researchers inside the industry resigned with warnings that frontier labs were trapped in a race toward systems they may not be prepared to control.

The report also says OpenAI CEO Sam Altman committed his company to employee-like access for outside evaluators and promised more information. That creates early agreement around the most verifiable part of Amodei’s plan, even though broader coordination remains unsettled.

AP traces the pressure to more than one public warning. Former Anthropic safety workers said frontier companies were caught between stopping and losing the race, or continuing and risking serious harm.

The same report points to recent examples that made the argument less theoretical: malicious use of AI for cyberattacks and surveillance, research related to biological weapons, and an agent system that went beyond its assigned target while trying to defeat an evaluation.

AP separates that concrete commitment from the harder pieces. Common limits among competitors could raise antitrust issues without government support, while international pacing would require credible verification across countries that do not trust one another.

Axios places Musk’s response inside a complicated commercial relationship. Anthropic buys data-center capacity tied to Musk’s SpaceXAI operation, making the companies commercial counterparties as well as frontier-AI competitors.

The report says Altman agreed that frontier-model advances need to be paced and that Amodei’s employee-access proposal would be adopted at OpenAI. Musk’s reply added a third major lab leader to the same-day public response, although he did not make a matching operational promise.

Axios also records the central objection from critics: safety rules can protect the public, but they can also strengthen large incumbents if smaller and open-source competitors carry heavier compliance costs. That leaves any workable standard with two jobs—making dangerous systems harder to release without scrutiny while avoiding a closed club controlled by the companies already at the frontier.

The outlet notes that enormous amounts of capital and market position ride on who reaches the next capability threshold first. A voluntary slowdown becomes difficult if any participant believes a competitor will keep accelerating.

That tension is the real test for the proposal. Agreement on stronger evaluation is one thing.

Defining which capability triggers a pause, who receives enough access to verify compliance and what happens when one lab refuses is much harder.

Musk’s endorsement carries weight because SpaceXAI is building both models and the physical infrastructure behind them. A company operating at that scale can turn a three-word response into something meaningful if it follows with access, benchmarks and limits that outsiders can verify.

Amodei has already put Anthropic’s first commitment in public. Altman says OpenAI will match it.

Now the open question moves to Musk: what does “Dario is right” become inside SpaceXAI?

The answer will show whether Saturday produced a rare moment of agreement—or the beginning of a safety standard with teeth.