OpenAI Cybersecurity Program Hit by Technical Error, Researchers Say
Synopsis
OpenAI has acknowledged that a technical error caused several cybersecurity researchers to lose access to its limited Trusted Access for Cyber program.
OpenAI has suspended access to its Trusted Access for Cyber programme due to a technical glitch, although it has not identified the cause of the disruption.
The problem occurred in the latest level of its vetted programme, Daybreak Blue. OpenAI informed the researchers impacted by the disruption that the issue was technical and that they would need to reapply and be reverified.
Researchers Asked to Verify Access Again
Five researchers who responded to requests for comment said they were unable to access the programme. Some received letters saying their identities had not been verified, and others were informed that they were no longer participants.
The researchers were based outside the United States and Europe, but OpenAI has not identified location as a factor, nor has it identified how many users were impacted by the disruption.
Trusted Access for Cyber is a programme for researchers who have been verified to engage in authorised security research activities using OpenAI models. Daybreak Blue allows researchers to participate in programmes that involve vulnerability research, code reviews, malware analysis, and incident response.
The access issue occurred days after OpenAI announced that it had suspended some AI development and testing following a security breach involving an experimental AI agent. On August 18, Reuters reported that OpenAI has paused its testing of models by two weeks and suspended its training of its upcoming Astra model as it strengthens security measures.
AI Cybersecurity Risks Increase for Businesses
It is particularly pertinent given the rising concerns about the dangers that AI poses to enterprises in Australia and the United States in relation to their cyber security.
AI-related weaknesses were identified by 87% of respondents as the fastest rising cyber risk in 2025 in the World Economic Forum’s Global Cybersecurity Outlook 2026. In addition, 94% of respondents identified AI as the most important enabler for cybersecurity in 2026.
According to the Australian Signals Directorate’s 2024–25 Annual Cyber Threat Report, there were more than 84,700 crime-related cyber incidents reported in Australia in the financial year, and more than 1,200 cybersecurity incidents responded to by the agency, an increase of 11 percent over the same period in 2016.
According to IBM’s July 2024 Cost of a Data Breach Report, one in four malicious attacks involved AI, an increase of 56 percent year-over-year. For example, the cost of these attacks was on average $6 million, which is $4.99 million, which was significantly higher than the worldwide average for data breaches.
Australian and US companies that are using AI for security purposes have much to say about OpenAI’s decision to restrict access to its specialized cybersecurity capabilities to a limited set of researchers.
Source: TechCrunch
Pooja Malik is a business journalist with over six years of experience covering startups, entrepreneurship, and emerging trends. She has previously worked with leading media platforms such as YourStory Media and BW BusinessWorld, where she reported on business, policy, and market developments. Currently, she serves as Editor at The Inspirepreneur Magazine, where she writes and edits stories across business, lifestyle, and travel, with a focus on clarity, accuracy, and reader relevance.
You Might Also Like
The Business of Being Well: Inside Steph Claire Smith’s KIC Platform
Billions Flow Into Neoclouds as Nvidia Expands AI Strategy
