The Access Layer Is Now a National Security Asset

For the first time, the U.S. government has used export-control authority to switch off a commercially deployed AI model. On June 12, 2026, the Commerce Department invoked the Export Control Reform Act to shut down Anthropic's Fable 5 and Mythos 5, and because the company could not verify nationality at its API, it had to go dark for every user on earth in a single directive.

What may appear as a regulatory act actually portends something much larger. This is a clear indicator that the world's most powerful governments fully recognize what AI can do, and it offers a signal that they plan to take action to control it. The public narrative centered on an alleged jailbreak technique that Amazon flagged. The message was that this constituted a legal trigger, but the real reason runs considerably deeper.

In a June 11 Senate Intelligence Committee briefing, Sen. Mark Warner said NSA Director Gen. Joshua Rudd had told him that Mythos, in an authorized red-team exercise, broke into almost all of the NSA's classified systems, not in weeks but in hours. The account, first reported by The Economist, has not been confirmed by any government agency, and the publication's editor later cautioned it should not be read literally. When that testimony landed, the question of whether Fable 5 should be available to any foreign national anywhere in the world became very simple, very fast.

The Industry's Identity Infrastructure Gap

The shutdown exposed a challenge that extends well beyond any single AI provider. Anthropic happened to be the company affected by this particular action, but the underlying issue exists across the frontier AI ecosystem. As AI capabilities advance, questions around access controls, identity assurance, governance, and regulatory compliance are becoming increasingly difficult to ignore.

More broadly, this introduces a new category of operational risk for enterprises building AI into long-term business processes. If access to a model can be suspended with little notice because governance or compliance requirements are not adequately addressed, organizations face uncertainty around the reliability of critical AI-dependent workflows. This is precisely why many security, fraud, identity, and governance professionals have been calling for stronger controls and clearer operating frameworks around frontier AI deployment.

Anthropic had no mechanism to selectively remove access by nationality. Being unable to verify user nationality in real time at API scale, the company had no option other than shutting Fable 5 down for every customer on earth. In a single directive, it initiated a total global blackout. Every enterprise customer, every developer, every API integration was gone simultaneously, because the identity infrastructure to do it cleanly did not exist. 

The most important question exposed by this event was surprisingly fundamental: who has access? In security and identity management, understanding who can access sensitive systems, under what conditions, and at what assurance level has long been considered a foundational control. Given the capabilities now emerging from frontier AI systems, it is not surprising that regulators and security professionals are increasingly focused on this question.

To be fair, the failure wasn’t because of an unexpected bug. It borne of a faulty design assumption that is shared by the entire industry. Anonymous, stateless API access was a feature, but it now turns out it was also a national security liability.

Identity Is Now the Compliance Primitive for Frontier AI

While the scientific community remains optimistic about AI's potential, the organizations building these systems are increasingly confronting a challenge that security professionals have dealt with for decades: powerful technologies require equally strong identity, security, and governance controls. The investment and attention devoted to advancing AI capabilities must be matched by investment in the infrastructure that governs access to them. Events like this illustrate what can happen when those two tracks evolve at different speeds.

Anthropic is first, not alone. It confirmed in June that it is using Persona to perform document and facial verification at the consumer layer, the first major lab to codify biometric identity checks for consumer access. OpenAI and Google do not yet require it, which tells you where the market is heading, not where it already is

With verified US citizenship data, Anthropic could open Fable 5 access to domestic users willing to submit an ID without requiring the Commerce Department to formally lift the directive. That is the deal being assembled right now. Identity verification will cease to be merely a compliance gesture and instead will become the mechanism by which AI capabilities get restored to paying customers. 

For fraud and risk leaders, this is the market signal: identity verification is becoming the toll gate for frontier AI. It will be a regulatory precondition for operating at all.

The Scale of the Threat That Got Us Here

It’s important to understand some context for what’s driving this in order to understand the opportunity it creates.

The Chinese lab extraction campaign is the most concrete illustration of what happens without identity controls at the access layer. Three Chinese research labs ran approximately 16 million exchanges through roughly 24,000 fake accounts. In a separate disclosure dated June 10, Anthropic told the Senate Banking Committee that an Alibaba-linked campaign ran more than 28.8 million exchanges through almost 25,000 accounts between April and June 2026, the largest such effort it has documented. It was a coordinated, systematic effort to distill Claude's capabilities into models that could be replicated and deployed without Anthropic's safety architecture. This might have looked like credential stuffing, but it was identity fraud in service of state-level AI capability transfer.

Every one of those 24,000 accounts was a fabricated identity that Anthropic's access layer accepted without challenge. The accounts looked real enough to pass whatever friction existed. The extraction ran until it was detected, but it was not prevented.

This is what identity failure looks like at national security scale. Synthetic accounts, operating at volume, used to strip proprietary AI capabilities and transfer them to adversaries. The access layer had no meaningful nationality filter, no behavioral signal that 24,000 accounts were running coordinated extraction, and no mechanism to catch it until after the damage was done.

That is the problem that now has the attention of the NSA, the Commerce Department, and the White House. At its core, this is first and foremost an identity management problem. Fraud, abuse, and policy violations are downstream consequences of not being able to confidently determine who is accessing powerful systems, under what authority, and at what level of assurance.

The Surveillance Backlash Is Coming

Here is the tension that will define this market for the next 24 months: The government wants identity at the AI access layer. The public is increasingly sensitive to how biometric data is collected, stored, protected, and shared, particularly when that information may be concentrated within a small number of providers and when transparency and privacy controls are not well understood.

AI identity verification is going to get dragged into the same fight. The moment "Anthropic has your government ID, your face scan, and a facial geometry template on file" becomes a news headline, and it will, you are going to see a significant consumer revolt against centralized identity verification for AI access. Third-party identity providers can help address operational and compliance requirements, but they do not eliminate broader concerns around data protection, privacy, governance, and consumer trust. Those concerns remain regardless of where sensitive identity information is ultimately stored.

This is precisely why consumer-controlled verifiable credentials exist. The correct answer to "prove you are a US citizen with legitimate use intent" is not "hand your ID and face scan to a VC-backed identity platform." The correct answer is a credential that lets users prove what they need to prove (nationality, age, fraud history) without surrendering raw biometric data to be stored, breached, or sold.

Passive IDV, identity verification that does not require users to actively submit documents or sit for a face scan, can provide better UX. But in this environment, it is the politically defensible alternative to building a surveillance infrastructure on top of AI access. The labs that get this right will have a meaningful advantage in consumer markets. The ones that build in ways that feel like surveillance will face regulatory and reputational exposure as the backlash matures.

The New Focus on Identity Isn’t So New

The Fable/Mythos shutdown is a proof point that identity infrastructure for AI has now become a policy conversation. The government has demonstrated it will shut down globally deployed AI models with 90 minutes of notice when it believes the access layer is inadequate. Every frontier lab now has to solve the same problem Anthropic is scrambling to solve: how do you verify the identity and nationality of users at API scale, with minimal friction, without building a surveillance apparatus that triggers a consumer revolt?

This is Prove's core capability. The specific attack vector, in this case, fake accounts used to extract AI capabilities at scale, is a variant of the synthetic identity fraud problem we have been solving in financial services for years. The detection signals are different, the regulatory context is new, but the underlying challenge is familiar: distinguishing real, authorized users from fabricated identities operating at volume with coordinated intent. Passive behavioral signals, device intelligence, document verification, and other risk-based identity controls all have an important role to play in solving this challenge. The goal should be to apply the appropriate level of identity assurance while minimizing unnecessary friction and preserving user privacy wherever possible.

The labs are moving. The government is forcing the timeline. The consumer backlash is coming. The window to be the identity infrastructure for AI access, rather than a vendor bolted on after the fact, is right now.