Anthropic-U.S. battle highlights AI power struggle
Add Axios as your preferred source to
see more of our stories on Google.
Photo illustration: Brendan Lynch/Axios. Photo: Evan Vucci/AFP via Getty Images
The U.S. government's move to stop Anthropic's Fable 5 and Mythos models turns a long-theoretical debate into a real-world test of who should have the power over world-changing AI.
Why it matters: Handing the reins to companies means trusting firms with huge commercial incentives. Leaving decisions up to governments could lead to secretive high-stakes decisions with little public approval.
Catch up quick: Anthropic was given 90 minutes last week to take down Fable and Mythos after Amazon raised concerns with U.S. officials about a jailbreak that could bypass Fable's guardrails and expose cybersecurity capabilities, sources familiar told Axios.
- The government imposed stringent export controls that ultimately led Anthropic to take the models offline entirely.
State of play: AI policy researchers and industry critics say the U.S. is now regulating frontier models through emergency intervention rather than a clear process.
- "Things feel very ad hoc," Adam Gleave, CEO at FAR.AI, an AI safety research firm, told Axios, adding "you're only as safe as the least safe model on the market."
- "In every sport, you have to separate the referee from the team players ... regulation should never come from industry," Connor Leahy, U.S. executive director of ControlAI, told Axios, arguing that government should have the authority on AI despite its current "capabilities."
Case in point: When Amazon flagged its concerns to the administration, it went to Treasury Secretary Scott Bessent, who was already close to the Mythos rollout because of its planned use by major U.S. banks.
- Model oversight has implications for national security, competition and free expression. And right now these decisions are being made through improvised channels, rather than a clear AI regulatory framework.
Between the lines: The most workable answer may be neither company self-policing nor unilateral government control, but an entirely new AI-fluent agency tasked with combining company testing, outside audits and government authority.
- "There needs to be a partnership between companies themselves and some kind of regulatory involvement to keep the companies honest," Patrick Wendell, Databricks co-founder, told Axios.
- Helen Toner, the head of Georgetown's Center for Security and Emerging Technology and a former OpenAI board member, agreed that "a good middle ground is to not put full control in either industry or government" and said what's needed is "a technically capable government agency that can engage deeply with companies on what they're building, how they're testing it, and how they're managing risks."
- Toner said a good first step is the set of state laws passed in New York, California and Illinois.
- Third-party audits combined with legal liability are also necessary, Toner says. She thinks the onus should be on the companies to ensure they're not taking unreasonable risks.
Zoom in: Toner also suggests moving away from the public release of models as the main safety checkpoint.
- "The companies have extremely strong commercial incentives to launch fast, so anything that has to happen pre-release (e.g. testing, government approval) will be under a lot of pressure to happen quickly — meaning it could easily turn out to be rushed," Toner said in an email.
Yes, but: Even models that haven't been released to the public are subject to cybersecurity risks.
- Toner said they could be misused internally or leak to adversaries.
- "A better approach would be to regularly (say, quarterly) assess the most capable models within each company."
The intrigue: Anthropic CEO Dario Amodei was calling for something close to this before the Fable/Mythos clash.
- Earlier this month he argued that frontier AI models should face mandatory technical testing and audits, and that the government should have the power to block or reverse deployment when third-party assessments show unacceptable risks.
Zoom out: The U.S. now has the power — and the willingness — to stop a frontier AI model from spreading.
- What it does not yet have is a clear answer for who should make that call, what evidence should trigger intervention and how the public will know the decision was justified.
What we're watching: Whether this becomes a one-off scramble — or the beginning of a more formal U.S. process for policing frontier AI models before and after release.
The bottom line: Neither companies nor governments are likely to accept the other having sole authority over frontier AI. Until there is a standard process for testing, disclosure, audits and intervention, the future of AI may be shaped less by rules than by who can get the government's attention first.
