Name: OpenAI's cyber model answers 95% of exploit prompts
Uploaded: 2026-08-11T09:00:00+00:00
Duration: 47 s
Description: OpenAI released GPT-5.6-Cyber on 10 August 2026, a model trained to find zero-days and build exploit chains. It completes 95% of advanced offensive-security requests that standard GPT-5.6 Sol completes 1.5% of the time, and it ships only through Daybreak Red, an application-vetted access tier.

OpenAI's cyber model answers 95% of exploit prompts.

Security · 2026-08-11 · 47s

OpenAI released GPT-5.6-Cyber on 10 August 2026, a model trained to find zero-days and build exploit chains. It completes 95% of advanced offensive-security requests that standard GPT-5.6 Sol completes 1.5% of the time, and it ships only through Daybreak Red, an application-vetted access tier.

Read the full story: OpenAI's Cyber Model Answers 95% of Exploit Prompts →

Transcript

OpenAI just shipped a model whose selling point is that it says yes. GPT-5.6-Cyber, announced August tenth, is trained to find zero days and build exploit chains. On OpenAI's own benchmark it completes ninety five percent of advanced offensive security requests. The public version of the same model completes one point five percent. Here is the detail everyone is skipping. Lifting the safeguards on the public model only moves it to two percent. The refusals were never the brake. The training was. And that training ships only through Daybreak Red, an application only tier that reads like a list of the biggest security vendors. Four days earlier, OpenAI paused a different model for being too good at this. So the safety control is no longer the model. It is the guest list.