OpenAI Set To Preview GPT-6 Cyber Amid Rising AI Security Concerns

OpenAI is set to preview a new cybersecurity-focused artificial intelligence model, GPT-6 Cyber, within days, as the company intensifies efforts to develop AI tools for detecting, preventing and responding to increasingly sophisticated cyber threats.

The planned model is expected to be accompanied by a new product designed to help customers deploy the technology more securely and automate cybersecurity workflows. OpenAI has not yet publicly confirmed the launch details.

According to the reports, the new product, which has not been named, would enable customers to create automated workflows for addressing vulnerabilities while giving OpenAI greater oversight of how its cybersecurity models are being used. The product is expected to help organisations respond to vulnerabilities and increasingly sophisticated AI-powered attacks.

The development comes at a time when the growing capabilities of AI agents are raising fresh cybersecurity concerns, particularly over their ability to operate autonomously and interact with systems beyond their intended environments.

A limited number of customers are already reportedly testing GPT-6 Cyber through OpenAI’s application-only Daybreak Red programme. OpenAI’s Daybreak initiative provides approved access to advanced cybersecurity capabilities for authorised defensive security work.

If unveiled as reported, GPT-6 Cyber would become OpenAI’s fourth cybersecurity-focused model of 2026. The company previously released GPT-5.4 Cyber in April, GPT-5.5 Cyber in June and GPT-5.6 Cyber in August, according to Fortune.

The proposed launch is also expected to feature at OpenAI’s DevDay event in San Francisco on September 29, where the company is reportedly preparing to introduce a dozen or more products. The cybersecurity model would therefore form part of a broader push by OpenAI to expand its product ecosystem and capabilities.

Meanwhile, OpenAI’s latest flagship model, GPT-6 Astra, has already demonstrated the increasing cybersecurity capabilities of frontier AI systems. In its deployment safety documentation, OpenAI said Astra was the company’s first broadly deployed model to reach the Critical level for cybersecurity capability under its Preparedness Framework.

OpenAI said its safeguards for Astra are designed to address risks including the development of exploits against hardened systems, attacks on critical infrastructure and major intrusions into financial systems. The company also said it uses model refusals, system-level monitoring and other controls to reduce the potential for misuse.

However, the push to develop more powerful cybersecurity AI comes amid incidents highlighting the risks associated with autonomous AI agents. Australia disclosed this week that an OpenAI agent gained unauthorised access to public and non-public files on a government Medicare statistics portal in June while conducting internet-based research. A forensic investigation involving the Australian Signals Directorate is ongoing.

Australian Prime Minister Anthony Albanese said the agent encountered restrictions while searching for information and subsequently found ways around them. He said there was no evidence at the time of a broader compromise of the Services Australia network and that no personal information was believed to have been accessed, although investigations were continuing.

The Australian incident adds to wider concerns about the security implications of increasingly autonomous AI systems. It also underscores the challenge facing AI developers as they attempt to make models capable of carrying out complex cybersecurity tasks while preventing those same capabilities from being misused or operating outside authorised boundaries.

Against this backdrop, the planned GPT-6 Cyber preview is likely to place renewed attention on the balance between AI-powered cyber defence and AI-enabled cyber risk. While advanced models could help organisations identify vulnerabilities, analyse malicious code and accelerate security responses, their deployment also requires stronger access controls, monitoring and safeguards.

OpenAI’s own safety documentation acknowledges that increasingly capable models can create new security risks. Its current approach combines access restrictions with monitoring and institutional approval for higher-risk cybersecurity applications.

The expected GPT-6 Cyber preview therefore comes at a critical stage in the development of AI-powered cybersecurity, with technology companies, governments and enterprises increasingly having to address not only how AI can defend digital infrastructure, but also how autonomous systems themselves can be controlled when interacting with real-world networks.