OpenAI opens cyber models to defenders

The Daybreak Cyber Partner program opens frontier models to companies that guard other people's systems: Accenture, IBM, PwC, NCC Group and others. Access grows, says OpenAI, and the safeguards grow with it.

In short

On August 10 OpenAI announced an expansion of the Daybreak Cyber Partner program: frontier cyber models go to security partners.
On the list: Accenture, IBM, Capgemini, Cognizant, EY, KPMG, PwC, NCC Group, SpecterOps. The goal: find vulnerabilities, judge which are a real threat, fix faster.
OpenAI's wording on access: more access, with the safeguards to match. The name GPT-5.6-Cyber appears nowhere in the announcement.

Checked on11 August 2026Responsible editorTsvetelin IvanovHow we workMethod · Corrections

I read the retelling first: OpenAI was supposedly shipping a cyber model with its safeguards stripped, all but built for writing exploits. Then I opened the original. No such sentence exists there.

The facts: on August 10 OpenAI announced an expansion of the Daybreak Cyber Partner program. Frontier cyber models go to security partners: Accenture, IBM, Capgemini, Cognizant, EY, KPMG, PwC, NCC Group, SpecterOps. The officially stated goal: finding vulnerabilities, judging which ones are a real threat, and fixing them faster, inside the systems defenders already use. The wording on access: more access, with the safeguards to match. There is no CVE: this is a model access program, not a software flaw.

Why a partner program instead of an API for everyone? Because this class of models is dual-use by birth. The same capability that finds the hole for the defender finds it for the attacker. A vetted door is the price of that knowledge.

And the gap between the aggregator headline and the official text is the gap between news and bait. One sells a click with fear, the other describes a partner program with audits. They are not the same story.

The partner list is also a list of the accountable. When access has names on it, there is someone to ask.

The original is always two clicks away. Open it.

The continuation OpenAI pledged $1 billion in access to its cyber models - our story from 3 September.

The continuation OpenAI gives Ukraine its vulnerability-hunting models - our story from 23 September.

The visual is generated code art. No third-party images.

OpenAI hands its cyber models to the defenders, with a gate and a list

More on this

Anthropic: China's GLM-5.3 builds working exploits almost like Mythos, and its safeguards fall to simple tricks

OpenAI's models built themselves a secret mailbox in the package server, and rode it into Hugging Face

Seven flaws in one day in the actively exploited catalogue, LiteLLM and Artifactory among them