OpenAI has announced a new cybersecurity model for authorized participants in Daybreak Red, the higher tier of its security program. The release comes as security professionals debate access to advanced models and the resources needed to handle their findings. OpenAI says it is devoting computing capacity to finding and patching weaknesses in open-source software.
OpenAI announced its new cybersecurity model on 5 November for approved participants in Daybreak Red. This is the higher tier of its security program, intended for authorized security research. The release places access within a managed program rather than opening it to every account. OpenAI’s system card identifies the model as GPT-6 Cyber and describes identity checks and organizational security requirements for users seeking access to it.[1], [2]
The launch accompanies a debate among security professionals about how advanced models are made available. Charge Robotics co-founder and technology chief Max Justicz told TechCrunch that Daybreak Blue, another level of the program, had recently become less useful for his research. He described having used an earlier model and reporting a finding to the Debian Project, which issued a patch. The reporting distinguishes the new model’s release from that earlier research experience.[1]
OpenAI says it wants to expand Daybreak to organizations of different sizes, independent researchers and open-source maintainers. The company has also announced support for participants protecting critical infrastructure, including electricity and water systems. Researcher Eric Wallace says OpenAI is allocating computing resources to finding and patching weaknesses in open-source software. TechCrunch’s account also describes the subscription and usage expenses borne by a researcher, putting practical resources alongside the question of model access.[1]
