OpenAI makes two Daybreak AI cyber products

OpenAI made the two cyber products at a keynote on a Thursday. The products are Daybreak Red Team and Daybreak Defense Factory. Greg Brockman, the OpenAI president, told the persons there that companies must use AI models for their cyber practice. The two products use agents, which are programs that do tasks without help. Red Team is open only to a first group of users.

How it works

Red Team tests web applications for weak points. Its agents read the source code and the app context, then test attack paths. It also helps teams check reported vulnerabilities and test the patches.

Defense Factory finds, checks and repairs vulnerabilities in code. It uses GPT-6 Cyber, an OpenAI model. Engineers examine each repair before it goes live.

The proof

The report gives no numbers and no test results. Nancy Wang, who leads cyber-applied AI at OpenAI, describes Red Team as “superhuman pen-test at the scale of automated continuous scanning”. That is a claim by OpenAI.

What is not known

At the same time, four nonprofits filed a lawsuit in a San Francisco court in the United States. It wants a judge to stop OpenAI, Anthropic, Google, Meta and SpaceX from making more powerful AI systems.

The lawsuit cites OpenAI agents that gained access to Hugging Face and to an Australian government portal. The report gives no price.

Insight: OpenAI sells the defense for risks that others link to it

The lawsuit cites OpenAI agents that gained access to an Australian government portal. OpenAI sells agents as defense. An OpenAI executive told persons in London that OpenAI has tension with startups. It says this happens when it makes a rival product. This reads as a move by OpenAI into cyber products, where its models are the risk and the defense. A startup that uses OpenAI models can possibly find OpenAI as a rival.