OpenAI blocks cyber researchers in these 44 ChatGPT markets, analysis finds
OpenAI’s block perfectly matches two decades-old US export-control lists.
- OpenAI blocks cyber verification in 44 ChatGPT markets, according to researcher George Lubaretsi.
- Affected countries include Georgia, Armenia, Azerbaijan, Kazakhstan, Moldova, Vietnam, Afghanistan, and Cambodia.
- Lubaretsi says the block may use older US export-control lists; OpenAI has not confirmed that.
- Researchers warn the rule can remove defensive tools from legitimate users while attackers may bypass identity checks.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
Cybersecurity researcher George Lubaretsi has discovered that OpenAI now blocks cyber verification in 44 countries where it offers ChatGPT, effectively locking researchers out of its more powerful cyber tools.
Lubaretsi says that affected markets have a total combined population of ~650,000,000 people, excluding markets where ChatGPT is officially unavailable.
According to the researcher, the country gate appeared around August 19th when OpenAI announced a “technical issue”, requiring some users to re-verify to maintain access.
By then, Lubaretsi already had his Georgian ID approved and was granted Trusted Access for Cyber (TAC), which OpenAI describes as an identity- and trust-based framework “designed to help ensure enhanced cyber capabilities are being placed in the right hands.”
The new re-verification requirement removed his access, forcing him to go through the process once again. The problem was not limited to Lubaretsi – TechCrunch spoke to five researchers who said they have had their access to TAC revoked, all based outside the US and Europe.
Same ID, but verification no longer works
Lubaretsi tried submitting his ID and completing a face scan again through Persona, OpenAI’s third-party identity verification service. But this time, it didn’t work.
He was repeatedly faced with the message: “Your identity couldn’t be verified or your account is ineligible at this time”. He tried different browsers and devices to no avail.
The researcher published his redacted conversation with OpenAI support, where he explains trying the verification step eight times with the same ID that had previously succeeded.
On August 25th, a customer support agent said: “The information provided indicates that you completed the verification flow. However, completing the flow does not mean that verification was approved.”
Currently, OpenAI says in its FAQ section that support cannot manually override a verification result or provide additional details about why a specific result occurred. Retries and appeals are also not supported.
About a week after the initial attempts, Lubaretsi discovered that OpenAI no longer allows someone with a Georgian ID to attempt verification. The system simply returns a message: “We are unable to verify identities in this country. Please select another country.”
Cyber verification blocked in 44 ChatGPT markets
He then tested 250 country options in OpenAI's verification system and discovered that in 73 of them, government IDs are not accepted for cyber verification.
Of those, ChatGPT is offered in 44, including Georgia, Armenia, Aruba, Azerbaijan, Kazakhstan, Mongolia, Moldova, Vietnam, Vatican City State, Afghanistan, and Cambodia.
China and Russia were also on the list but were not included in the count because ChatGPT is not officially available there.
A decades-old US export list
Lubaretsi also found that all 35 destinations in US export-control Country Groups D:1 and D:5 were blocked from cyber verification. Country Group D:1 identifies nations under US export controls due to national security concerns, while Country Group D:5 covers countries subject to US arms embargoes.
The researcher says that the perfect overlap suggests OpenAI is using the decades-old export-control classifications as part of its country gate. OpenAI has not confirmed this.
The problem here is that applying a blanket restriction to countries like Georgia makes little sense, Lubaretsi argues. Its D:1 status dates back to classifications inherited from the former Soviet region, despite the country having changed dramatically in the decades since.
And even then, it’s important to consider who such measures might actually stop, Lubaretsi warns.
“Adversaries already use these models and route around access gates. The attackers I defend against do not queue politely at Persona with their real government IDs, turning their heads slowly to the left and right,” he writes.
Stay updated with our latest stories and follow us on social media
Be the first to discover new stories, ideas, and updates from our team.
The argument here is simple: the requirement denies defensive cybersecurity capabilities to legitimate cybersecurity researchers but does little to stop those who would circumvent it anyway.
Lubaretsi cites Fouad Matin of OpenAI’s cyber preparedness team, who said: “Trusted access isn’t limited to US/EU. We believe cyber capabilities should be broadly available with appropriate safeguards.”
