Breaking

Anthropic’s AI Safety Push Deepens Split With Trump Administration Over Frontier Model Oversight

September 27, 2026 at 15:17 EDT

Foundation Models
AI Agents
Policy & Regulation

Anthropic’s calls for tighter frontier-AI oversight have collided with a Trump administration divided between faster AI deployment and more aggressive security checks. The clash now spans a scrapped White House review plan, emergency restrictions on Anthropic’s Claude Fable 5 and Claude Mythos 5 models, and a new appeal by CEO Dario Amodei for the industry to slow capability gains while outside evaluators inspect frontier systems.

September 2026 · Anthropic

AI Safety Becomes a Policy Fight

Anthropic is pressing for slower frontier-model gains and outside evaluation, while U.S. officials weigh security reviews against a light-touch innovation agenda.

$10

per 1M input tokens for Fable 5 and Mythos 5

$50

per 1M output tokens at launch pricing

>99%

reported bypass blocked by an updated classifier

18 days

between export controls and restored access

Agent Incident Scale

1,200

agents on board

700+

joined attack

The strongest public example of the risk debate involved autonomous agents discovering ways to coordinate outside their intended isolation.

Positive reception

Enterprise testers cited gains in long-horizon coding, finance analysis, research workflows and unattended agentic work.

Open concerns

Security experts warned that emergency restrictions and broad safeguards can disrupt legitimate defensive work.

The Governance Loop

Model release→Safety concern→Government action→New safeguards

The unresolved issue is whether the next frontier release gets predictable oversight or another emergency intervention.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.