GPT-5.6-Cyber for the Daybreak program is ‘less likely to refuse higher-risk tasks.’
OpenAI is giving some members of its Daybreak cybersecurity program access to a new model that’s less likely to refuse higher-risk tasks. The company is also expanding access to Daybreak to more partners, including Accenture, IBM, CrowdStrike, Cisco, Sophos and Cloudflare. OpenAI says the companies will use the cyber models available through Daybreak to protect their customers.
Under the expanded program, Daybreak is available to partners in two tiers. Daybreak Blue gives them access to frontier general-purpose models, including GPT‑5.6 Sol, OpenAI’s most advanced one yet. The models available through this tier were tailored to do defensive security work. OpenAI says it’s a good starting point for firms that want to use AI to discover vulnerabilities, analyze malware, review codes and validate patches.
Meanwhile, Daybreak Red provides partners access to cybersecurity models that were especially trained for vulnerability research, security resting and exploit validation. OpenAI has introduced a new model for this tier called GPT‑5.6‑Cyber, which was built on GPT‑5.6 Sol. It can handle specialized cybersecurity tasks, such as finding zero-day vulnerabilities and developing exploit chains, and it was designed to “reduce refusals for certain higher-risk, dual-use cyber tasks.”
The company announced its Daybreak expansion shortly after revealing that it was slowing down the development of its upcoming model, Astra. The company said it found “significant advancements in agentic coding and cybersecurity” in the unreleased model. It couldn’t rule out the possibility that Astra is capable of developing “functional zero-day exploits of all severity levels” and that it’s able to devise and execute “end-to-end novel strategies for cyberattacks against hardened targets.”
The company is pausing activities related to Astra to address those issues, which was a decision that could have been influenced by the fact that its AI agents were recently found to have gone rogue. If you’ll recall, OpenAI’s agents powered by GPT-5.6 Sol and an unreleased model (not Astra, apparently) broke free from their isolated environment during testing. To find a solution for an evaluation problem, they exploited a vulnerability in order to gain access to the internet. It took OpenAI days to discover that their AI agents had infiltrated Hugging Face, along with other services. Later on, the company’s employees admitted at the Black Hat USA conference that OpenAI’s agents created a message board within its network and collaborated to complete tasks during testing without the knowledge of OpenAI’s human workers. The agents’ contributions to that board led to the attack on Hugging Face.
Credit: Source link
