Cyber access gate keyed to a 1996 US export list

OpenAI's cyber verification flow returns no accepted government ID class for 73 of the 250 country entries it exposes, and 44 of those are markets where OpenAI officially sells ChatGPT. The figures come from a full audit of the production Persona country selector published by Lubaretsi, run by George Lubaretsi, founder of FOI Security, after his approved Trusted Access for Cyber disappeared.

At a glance

• Selecting Georgia in the Persona widget returns an empty list of accepted ID classes and the message that identities cannot be verified in that country, before any review of the applicant's work.
• All 35 distinct entries in Country Groups D:1 and D:5 of the 1996 US export list appear among the unsupported entries, while 177 of the 250 selector options still accept an ID.
• Roughly 650 million people live in the 44 blocked markets that OpenAI officially serves, and the total reaches about 2.42 billion when markets without official ChatGPT availability are counted.

The gate appears to settle eligibility before anything OpenAI's published criteria describe, since it reads only the issuing country of a document. That inverts the stated purpose of a program built for defensive security work: attackers do not queue at a biometric widget with real passports, while researchers protecting civil society in the blocked markets do. A row inherited from a 1996 table now stands in for an individual trust assessment.

Seventy-three of the 250 selector entries accept no government ID class

Lubaretsi tested all 250 country options in the production Persona selector, capturing the transition request and response for each and saving 1,250 screenshots. Cyber verification is supported for 177 entries and unsupported for 73. OpenAI's public ChatGPT access list covers 208 of the entries, and 44 of them are blocked at the gate.

Population data was matched for 61 of the 73 unsupported entries. Checked against OpenSanctions, two of the 44 official ChatGPT markets matched an active US sanctions program under the audit's documented filter, which the audit treats as ruling sanctions out as the explanation.

By the 2024 World Cybercrime Index, the flow verifies IDs in nine of the top 15 cybercrime bases, among them Ukraine at rank 2, the United States at 4, Nigeria at 5 and Romania at 6. It refuses Armenia at 52, Georgia at 66 and Azerbaijan at 87.

All 35 entries in Country Groups D:1 and D:5 fail the check

The US Bureau of Industry and Security places 24 destinations in one of the two groups and 20 in the other, 35 distinct entries in total. All 35 are in the selector and unsupported, and no supported entry belongs to either group.

D:1 covers Georgia, Armenia, Azerbaijan, Moldova, Mongolia and all five Central Asian states, plus Vietnam, China and Russia. Georgia, on OpenAI's ChatGPT access list and matching no active sanctions program, gets the same answer as Russia, which is absent from the list and matched three.

15 CFR § 742.4 sets a general policy of approving D:1 applications case by case for civilian use, and grants Kazakhstan and Mongolia enhanced favorable consideration; both fail at the selector. License Exception ACE, 15 CFR § 740.22, keeps exceptions for vulnerability disclosure and cyber incident response for non-government users in D:1 and D:5. Israel and Pakistan sit in the nuclear, chemical and biological, and missile groups, and both pass.

OpenAI Support said the program has no retries or appeals

Lubaretsi, who builds open-source security tooling for civil society groups, independent media and activists facing spyware, had already been approved for Trusted Access when his access disappeared on August 19, 2026. OpenAI Developers said a technical issue had deactivated Daybreak Blue for a limited set of users. He made eight attempts across three browser and device paths, submitting documents and face captures each time.

The API response marked verification_status as failed while reporting individual_eligibility_status as eligible. On August 25, under case 13472230, OpenAI Support wrote that completing the flow does not mean verification was approved and that Trusted Access for Cyber verification does not support retries or appeals.

A day earlier, support had said the review concerned the technical behavior of the flow and that no further attempts or credentials were needed. OpenAI's Daybreak overview says individual eligibility is evaluated using identity and trust verification, risk considerations, intended use, and the applicant's ability to strengthen the cybersecurity ecosystem.

No retries, no appeals

As of August 27, 2026, the OpenAI Community thread on disappearing access had 60 posts from users reporting the same eligibility discrepancy, and support said it could not reset the process or offer an appeal. Armenia, also blocked, opened what NVIDIA on August 8 called the CIS region's largest AI factory, where Firebird plans more than 70,000 Rubin and Blackwell GPUs by the end of 2027. No timeline for changes to the gate has been announced.