US Order Forces Anthropic to Disable Fable 5 and Mythos 5 for All Foreign Nationals
In an unprecedented move, the US government ordered Anthropic to block every foreign national from accessing its two most capable models, Fable 5 and Mythos 5, citing national security. It’s the first export-control measure to target specific AI models — and according to Anthropic, the directive landed at 5:21pm Eastern on Friday and extends even to the company’s own foreign-born employees. To comply, Anthropic abruptly disabled both models for all customers; every other Claude model stayed online.
The concern is specific: Mythos is unusually good at finding software vulnerabilities that have sat undiscovered for decades — a capability already used by US authorities and select firms to patch holes, but one that could become a cyberweapon in the wrong hands. The government believes Fable 5’s safeguard against that use can be jailbroken. Anthropic calls it a “misunderstanding,” warning the same standard applied across the industry would “essentially halt all new model deployments for all frontier model providers.”
Trump Orders Federal Agencies to Stop Using Anthropic Technology
Running alongside the export order is a second, broader escalation: the administration directed US government agencies to stop using Anthropic technology, in what reporting framed as a clash over AI safety. Together the two moves put Anthropic in an extraordinary position — simultaneously deemed too sensitive to share abroad and, on the federal side, told to stand down at home.
For a lab that has built its brand on working closely with government on safety, that’s a sharp reversal. The details are still developing, but the signal is clear: the relationship between the most safety-vocal frontier lab and Washington has entered a far more adversarial phase.
Sonnet 4 and Opus 4 Retire Monday — Migrate Before June 15
The clock that’s been ticking all week runs out Monday: Claude Sonnet 4 and Claude Opus 4 are scheduled for retirement on June 15. After that, calls to those model IDs return errors. The fix is a one-liner — point your code at claude-sonnet-4-6 and claude-opus-4-8 respectively — but it’s a one-liner that silently breaks production apps if it’s missed over the weekend.
If you run anything against the older IDs, this is the weekend to grep your configs. The successors are drop-in for most workloads, but it’s worth a quick regression pass on prompts that were tuned tightly to the 4-series behavior.
June 15 Billing Change: Agent SDK and claude -p Move to Separate Credits
Also landing Monday: Anthropic moves the Claude Agent SDK, claude -p, Claude Code GitHub Actions, and third-party agents off your subscription limit and onto a separate monthly credit pool — $20 on Pro, $100 on Max 5x, $200 on Max 20x. Interactive Claude Code usage stays on your plan; it’s the headless and automated paths that change.
If you’ve been running scheduled jobs or CI agents on a subscription seat, budget accordingly — the metering, not just the model, is changing under those workflows.
Advisor Tool Hits Public Beta; Claude Code Gets Nested Sub-Agents
On the platform side, Anthropic launched the advisor tool in public beta: pair a fast executor model with a higher-intelligence advisor that offers strategic guidance mid-generation, aimed squarely at long-horizon agentic workloads. Managed Agents also picked up scheduled deployments, vault-stored credentials, and richer session-thread webhook events.
Claude Code, meanwhile, added nested sub-agents, smarter model and region handling, a new plugin search, and tighter Chrome, VS Code, and terminal workflows. The throughline is orchestration — the tooling is increasingly built to run many agents in the background rather than one chat in the foreground.
Anthropic Files Confidentially for an IPO
Anthropic confidentially filed for an initial public offering on June 1, setting up what could be the first true public-market test of the AI boom. A confidential filing lets the company start the SEC process without disclosing financials yet, but it’s a clear signal of intent — and a marker of how fast the lab has scaled.
The timing is striking against this weekend’s headlines: the same models drawing national-security scrutiny are the ones underpinning the growth story Anthropic will eventually have to put in front of public investors. Regulatory risk just moved from a footnote to a front-page item.
20+ New Legal Connectors and 12 Practice-Area Plugins for Claude
On the ecosystem front, Anthropic released more than 20 new legal MCP connectors and 12 practice-area plugins, expanding how law firms and in-house teams use Claude across research, contracts, discovery, matter management, and legal aid. It’s a concrete example of the vertical-by-vertical playbook: bundle the connectors and workflows a profession actually uses, rather than leaving every firm to wire up MCP from scratch.
Paired with Managed Agents now able to run in a customer-controlled sandbox and connect to private MCP servers, the message to regulated industries is consistent — bring Claude to your data, on your infrastructure.
The Capability That Sold Mythos Is the One That Got It Pulled
There’s a hard irony in this weekend’s order. The thing that made Mythos valuable — finding decades-old software vulnerabilities faster than any human team — is exactly the thing that got it classified as a national-security risk. Anthropic spent months arguing these models were powerful enough to be dangerous. Washington appears to have agreed, and then acted on it more aggressively than the lab expected. Be careful what you successfully warn about.
The deeper problem is precedent. If “a safeguard might be jailbreakable” is enough to bar a model from foreign access — including a company’s own employees — then, as Anthropic notes, no frontier model clears the bar, because none are provably unjailbreakable. That’s not a Fable 5 problem; it’s an industry one. How this single order gets generalized will shape who can ship what, and to whom, for years.
An IPO Into a Headwind
Filing confidentially gives Anthropic room to wait out the noise before it shows numbers — and it may need every bit of that room. A public offering rests on a clean growth narrative, and this weekend just complicated it: a federal agency ban dents domestic enterprise demand, and a foreign-access block caps the international upside of its flagship models. Neither is necessarily permanent, but both are now things the S-1 will have to address head-on.
The optimistic read is that regulatory entanglement is a sign of how central Anthropic has become — you don’t get export-controlled for being irrelevant. The bearish read is simpler: the company is testing public markets at the exact moment its relationship with its home government is at its rockiest. Both can be true, and investors will price the tension.