On June 9, Anthropic launched two new flagship Claude models. Fable 5 and Mythos 5.
On June 12, the US Commerce Department issued a directive citing a narrow national-security-flagged jailbreak. Foreign nationals had to lose access. Anthropic could not surgically gate by nationality, so on June 13 they disabled both models worldwide. Their own foreign-national employees lost access to their employer's product. Microsoft banned internal use over the new data-retention terms.
24 hours from launch to lights-out.
This is not a post about whether Anthropic made the right call. They were threading a government directive against an installed user base inside a 24-hour window. There were no clean choices.
This is a post about what the rest of us should be thinking when we read the story.
Identity-mediated services live or die by identity decisions
Fable 5 is a model. But what happened to it was not a model failure. It was an access-control event. A government directive defined a class of people who could not have access. The provider could not filter at that level. So the whole service went dark.
Every cloud-mediated service has this same shape.
- Your Okta tenant lives until your identity vendor decides it does not.
- Your AWS account lives until AWS decides it does not.
- Your Stripe account lives until your underwriter decides it does not.
- Your Anthropic API key lives until policy or directive decides it does not.
We work on this category every day at Butterfly. Most teams have not internalized how thin the line is.
The directive is rare. The dynamic is not.
The Commerce directive was unusual. The dynamic underneath was not.
In a typical month we see:
- Customers locked out of their own Okta tenant because the only admin was off-boarded without a recovery account in place.
- Customers who pushed a network-zone rule that locked themselves out.
- Customers who lost their Workforce SAML cert mid-window and could not bring it back from a backup because there was no backup.
- Customers whose Workflows fired against deleted apps for a week before anyone noticed.
None of those events made the news. All of them looked, to the affected team, exactly like Fable going dark. Service is there. Service is not there. Identity that mediates access is the choke point.
What a real recovery story looks like
If a directive went out today telling Okta to revoke a class of identities tomorrow, what would your week look like?
If your answer is "we would open a P0 with our IDP and hope," you do not have a recovery story. You have an absence of one.
A real recovery story is:
- Point-in-time backups of every Okta object, every day, encrypted, with retention you control.
- Restore preview before commit. Diff against any prior backup. Know the blast radius before you press the button.
- Warm-standby Continuity so your workforce stays signed in while the primary is being put back together.
- Signed audit evidence to hand to your auditor, your regulator, or your board on the bad day.
- A Restore Readiness Score so you find out you cannot recover during a tabletop, not during the actual fire.
We built Butterfly for this. The free Chrome extension takes a 60-second snapshot of your Okta config with no signup, and tells you what your recovery posture actually looks like. chrome.google.com/extension/butterfly-for-okta
What we are not saying
We are not saying Anthropic acted wrong. We are not saying the directive was wrong. We are not selling a "cloud is unsafe, come on-prem" pitch.
We are saying: every team that depends on an identity-mediated service should assume that service can go dark in 24 hours. Not because the provider is bad. Because the world is shaped that way now.
Have a recovery story.
If you want to talk about yours, grab 30 minutes: calendly.com/mick-butterflysecurity/30min