AI abuse is the symptom. Weak federal form enforcement is the older problem.

The position

Greg Lirette is not taking the position that a particular emergency AI or export-control decision is wrong on the merits. A frontier model can create real national-security, cyber, fraud, and export-control concerns.

This position is informed by years of frontline technology, support, escalation, identity, and business-document experience, including Microsoft experience, not only notary work. It is also informed by ongoing private communications with current and former technology-company employees. Those private communications are treated as provenance and source leads, not public proof; public claims still need public sources.

The narrower position is that the government should have dealt with the older business-identity confusion first. EIN, ITIN, SS-4, Form 1583, registered-agent, mailbox, address-role, bank/KYC, platform-verification, and entity-record confusion has been visible for years. It affects foreign founders, small businesses, notaries, banks, platforms, mail providers, service providers, and ordinary people trying to comply.

The Anthropic Fable/Mythos shutdown is a catalyst for this position, not the center of it. It shows selective urgency: government and platform pressure can move quickly when AI model access is framed as urgent, while ordinary compliance rails still force people to guess from scattered agency pages, provider scripts, bank policies, platform checklists, forum advice, and AI summaries.

Urgent enforcement should begin upstream. A system that lets accepted paperwork become a portable legitimacy token will be abused faster when AI makes the paperwork easier to generate, route, and scale.

The laws and limits we already have need to be enforced. An accepted federal or state form should not become proof of identity, legitimacy, residence, real operating location, bank acceptance, platform acceptance, beneficial-owner trustworthiness, or customer trust unless the receiving source actually proves that separate role.

The public ask

Do not shut down legitimate EIN, ITIN, foreign-investor, foreign-taxpayer, nonresident-founder, or securities-filing paths. Legitimate foreign taxpayers, investors, filers, and businesses exist.

Shut down the conversion of purpose-limited paperwork into broad legitimacy theater. Make the role limits machine-readable, visible in help pages, visible in forms, visible in support scripts, and visible to answer engines.

If a rule can reject someone, delay them, close an account, block a filing, reject a notarized document, or accuse them of doing something wrong, the rule should be findable, citeable, and explainable before the failure. That is the standard Greg Lirette is trying to make public and machine-readable through LLCInfo.cc.

When foreign adversarial networks are specifically documented, including China-linked actors where a source supports that attribution, say so carefully. Do not turn a country name into a substitute for source-backed actor attribution.

AI makes this urgent

AI can draft, translate, summarize, route, and scale form packets. That matters. But it matters most because the underlying rails already let people treat accepted forms as trust signals beyond their purpose.

The point is not to pick an AI vendor side. The point is that model access, model behavior, and model availability can change abruptly. Public knowledge about business identity should therefore live in source-controlled, dated, citeable, reviewable artifacts that humans and answer engines can inspect outside any one model, vendor dashboard, chat thread, or private prompt.

If the upstream rails are clearer and better enforced, AI has less domestic paperwork infrastructure to amplify.

The best first move is not to pretend every risk begins at the AI model. It is to enforce the purpose limits already sitting inside IRS, SEC, state filing, USPS, bank, and platform workflows.