Anthropic removes Fable 5 and Mythos 5 access after US directive

• Anthropic said a US government directive forced it to suspend access to both Fable 5 and Mythos 5, and ClaudeDevs' suspension post says active Fable 5 sessions now error out while new sessions fall back to a default model or Opus 4.8.
• The blast radius reached partner tools immediately: Linear's failover notice says Linear Agent automatically switched affected coding sessions to Opus 4.8, and ClaudeDevs' Build Day update says Anthropic's own Fable 5 hack event had to run on Opus 4.8 the next day.
• The shutdown landed days after Anthropic pitched Fable 5 as a public Mythos-class model with automatic safety reroutes to Opus 4.8 on flagged bio and cyber prompts, according to the Fable page, while ClaudeDevs' safeguards apology had already acknowledged that some Fable interventions were too opaque.
• Usage caps were reset after the outage, with ClaudeDevs' rate-limit reset post announcing fresh 5-hour and weekly quotas for all users.
• Community reaction focused less on benchmark loss than on how abruptly access vanished, with a widely shared foreign-nationality complaint zeroing in on enforcement and an open-source argument from LLMJunky treating the incident as proof that frontier access can disappear overnight.

You can read Anthropic's full statement on the suspension, compare it with the still-live Fable product page, and see how awkward the timing was next to Anthropic's own policy essay, published two days earlier, which argued government should have power to block dangerous frontier deployments. Even the partner fallout surfaced fast: The Information's report says Anthropic blindsided business partners, while a Hacker News thread quickly filled with reports of Fable prompts already downgrading to Opus 4.8 before the full cutoff.

Anthropic's official statement says the government directive arrived at 5:21 p.m. ET on June 12 and applied to any foreign national, inside or outside the US, including Anthropic's own foreign national employees. Anthropic says that scope forced it to disable both Fable 5 and Mythos 5 for everyone, not just the people named in the order.

The public product impact was blunt. According to ClaudeDevs' suspension post, existing Fable 5 sessions end with an error, new Claude sessions run on a selected default model or Opus 4.8, and API requests to Fable 5 fail until developers swap integrations to another model.

The fastest secondary effect was failover. Linear's failover notice says workspaces that had enabled Fable 5 in Linear Agent were automatically switched to Opus 4.8 so coding sessions could keep running.

Anthropic had to do the same thing in public. ClaudeDevs' Build Day update says the company's June 13 Fable 5 Build Day in San Francisco still happened, but on Opus 4.8 instead of the model the event was built around.

The scramble fits The Information's reporting, which says partners were not warned in advance. Anthropic also tried to soften the immediate usage shock by resetting quotas, and ClaudeDevs' rate-limit reset post says 5-hour and weekly rate limits were refreshed for all users a little after the suspension.

The odd part is that Fable 5 had already been shipping with fallback logic before the government stepped in. Anthropic's Fable page says flagged bio and cybersecurity queries automatically route to Opus 4.8, and that API customers need fallback handling in place.

One day before the shutdown, ClaudeDevs' safeguards apology said Anthropic was changing another class of interventions, frontier-LLM-development safeguards, from invisible degradation to visible fallback. The post says Anthropic picked invisible safeguards to ship faster with fewer false positives, then reversed course after backlash.

That earlier controversy shows up in user reports. danshipper's safeguard-trigger post says a longer Fable project fell back to 4.8 after about 10 minutes, while a Hacker News discussion includes similar complaints about prompts getting safety-flagged and downgraded before people could really test the model.

Anthropic's statement is framed around a jailbreak concern, but the public conversation locked onto the scope of the order. The company says the directive covered foreign nationals everywhere and included Anthropic employees, which is why the broad shutdown read as much like an access-policy shock as a safety incident.

Reaction clustered around three concrete fears:

• Enforcement: the most shared scope complaint asked how a ban on "foreign persons inside the US" could be enforced cleanly.
• Identity gates: petergyang's follow-up predicted ID verification may become a requirement for top-tier model access.
• Dependence risk: LLMJunky's post about the faucet switching off argued the episode showed how quickly a closed model can disappear from active workflows.

The creative crowd felt the whiplash too. higgsfield_ai's scriptwriting demo had been showing off Fable 5 for screenplay-style work two days earlier, and danshipper's backlog joke captured the mood after the cutoff: there was briefly a model people felt could one-shot much more of the work.

The sharpest extra detail sits outside the tweet storm. In Anthropic's June 10 policy essay, Policy on the AI Exponential, the company argued government should have legal authority to "block or deter" dangerous frontier deployments. Two days later, Anthropic became the first obvious test case for that idea.

Anthropic's own statement says it disagrees with this specific order and calls the disclosed jailbreak concerns minor and previously known. But the sequence is still striking: a company publicly lobbying for stronger frontier-AI stop powers on Tuesday, then abruptly pulling its newest Mythos-class models on Thursday night after the government used one.