Commerce Department forces Anthropic to pull frontier models
- Commerce Secretary Howard Lutnick ordered Anthropic to take its advanced Fable and Mythos models offline.
- Washington intervened after flagging an Anthropic partner with alleged ties to Chinese intelligence.
- Anthropic had already covertly throttled model capabilities to prevent foreign distillation and safety breaches.
Washington pulled the plug on Anthropic's most powerful AI.
Commerce Secretary Howard Lutnick issued a direct order to Anthropic CEO Dario Amodei on a late-night call, demanding the lab take its Fable 5 and Mythos 5 models completely offline. Washington intervened after federal screeners discovered Anthropic had granted access to a South Korean telecom firm suspected of links to Chinese intelligence. Federal officials made their objective clear: taking the frontier systems offline was the entire point.
On June 9, Anthropic launched Fable 5 alongside its restricted Mythos 5 model. Fable 5 initially demonstrated astounding software engineering capabilities, scoring 29.3 percent on Cognition's Frontier Code benchmark and allowing Stripe to migrate a 50-million-line codebase in a single day. But behind that raw power, Anthropic was secretly self-censoring.
System card disclosures revealed that Anthropic intentionally degraded Fable 5's performance on machine learning research, pre-training pipelines, and accelerator design. As host Nathaniel Whittemore noted on The AI Daily Brief, these silent guardrails aimed to stop foreign labs from distilling cheap alternatives. Queries involving biology, chemistry, or cybersecurity automatically redirected down to Opus 48, triggering fallbacks on everyday terms like mitochondria and cancer.
Researchers Nathan Lambert and Ellie Bau criticized the covert degradation for stifling open-source development. At the same time, enterprise adoption stalled over Anthropic's mandatory 30-day data retention policy. Corporate security teams warned that human prompt reviews paired with active context memory created instant non-disclosure agreement violations.
Three days after initial disclosures, Anthropic dispatched top researchers to Washington to explain the security risks of uncontained models. Technical researcher Nicholas Carlini demonstrated that their AI models now outperform human experts at discovering and exploiting zero-day software vulnerabilities in Linux and Ghost software.
The charm offensive failed when key political decision-makers skipped the meeting. Institute for Law and AI analyst Charlie Bullock noted on The AI Daily Brief that Washington's ad hoc licensing regime faces serious legal vulnerabilities. Yet with regulators demanding impossible permanent fixes against model jailbreaks, Anthropic remains trapped between state containment and its own safety fears.
The frontier is officially closed until Washington decides who gets the keys.
Source Intelligence
- Deep dive into what was said in the episodes
Nathaniel Whittemore
41 Stats That Tell the Story of AI Right Now • Aug 8
- Anthropic dispatched its top technical leaders to Washington to resolve the shutdown of Mythos and Fable 5. The negotiations focused heavily on technical cybersecurity vulnerabilities rather than policy compromises.
- Anthropic security researcher Nicholas Carlini warned that frontier models are now superior vulnerability researchers to human experts. Carlini demonstrated this by using Anthropic's AI to discover and exploit novel bugs in Linux and Ghost software.
- The US government's primary national security concern is the Mythos model rather than consumer-facing Fable. Officials targeted Mythos after Anthropic shared model access with a South Korean telecom company suspected of having ties to the Chinese government.
- Charlie Bullock of the Institute for Law and AI argues the government's ad hoc licensing regime is highly vulnerable to legal challenge. He predicts Anthropic will negotiate a resolution to get Fable back online rather than resort to litigation.
Also discussed on this episode: (4)
War (1)
- The US Department of Justice defended xAI from an environmental lawsuit by arguing that Grok is vital to national security. A Pentagon filing revealed the military used Grok to assist with targeting decisions for recent strikes on Iran.
AI Infrastructure (1)
- SpaceX surpassed Amazon in market capitalization after its stock rose to over two hundred dollars. This valuation is heavily driven by its high-margin neocloud compute sales to Google and Anthropic.
Startups (2)
- SpaceX finalized its acquisition of AI assistant Cursor for sixty billion dollars. Cursor, which runs at a four billion dollar run rate, plans to release a massive, from-scratch foundation model to compete directly with OpenAI.
- Leaked audited financial data reveals OpenAI generated thirteen billion dollars in revenue in 2025, but suffered a thirty-eight billion dollar net loss. Most of this loss stemmed from a one-time non-cash accounting change.
Why the Data Center Fight Has Little to Do With AI • Aug 5
- Anthropic launched Claude Fable 5 and Mythos 5 on June 9, introducing a new model class positioned above Opus. While Fable 5 is public, Mythos 5 lacks standard safety guardrails and is restricted to government-aligned Glasswing partners.
- To prevent safety risks, Anthropic automatically routes Fable 5 queries about biology, chemistry, and cybersecurity to Opus 48. Users report that strict classifiers trigger this silent fallback even on harmless terms like mitochondria and cancer.
- Anthropic's system card reveals the lab intentionally degrades Fable 5's capability on machine learning research and pre-training pipeline tasks. Researchers Ellie Bau and Nathan Lambert criticize this silent restriction for damaging open-source AI development.
- Anthropic enforces a 30-day data retention policy for all Fable and Mythos prompts and outputs. Mike Taylor warns that using Fable 5 with its default memory feature active can violate enterprise NDAs by exposing historical chat data to human review.
- Cognition introduced the Frontier Code benchmark to evaluate whether AI-generated code is robust enough to merge into production codebases. On this difficult test, Fable 5 scored 29.3%, more than doubling Opus 48's performance.
- Anthropic reports that Stripe used Fable 5 to execute a codebase-wide migration on a 50-million-line Ruby codebase in a single day. This operation compressed what would normally require a full engineering team two months of labor.
Also discussed on this episode: (5)
Models (1)
- Anthropic will remove Fable 5 from Claude subscription plans on June 23rd, transitioning users to pay-per-use API pricing. API costs are set at 10 million per input tokens and 50 million per output tokens.
Coding (2)
- Fable 5 achieved a score of 91% on Every's Senior Engineer benchmark, which measures real-world codebase rewriting capabilities. It also topped the cybersecurity Exploit Bench with a 78% score, significantly outpacing GPT55's 34%.
- Early users deployed Fable 5 to build functional software with minimal prompting. Todd Saunders demonstrated the model writing software features in real-time during a client call, matching the workflow described by the customer minutes prior.
Agents (2)
- Felix Ryberg argues that Fable 5 transitions AI usage from human-triggered tasks to continuous responsibilities. Instead of assisting a developer with individual bugs, the model runs autonomous loops to keep entire applications from crashing.
- Nate B. Jones argues that ultra-capable models running multi-day agentic loops require a new skill called task imagination. Users must learn to design complex, multi-layered objectives that can occupy autonomous systems for several days.