← Shelf service.News

GPT-6 Cyber launches for vetted defenders as OpenAI raises bug bounties

OpenAI released GPT-6 Cyber to a small group of vetted Daybreak Red customers. Its researchers used the model to find 81 previously unknown high-severity vulnerabilities in Firefox, which Mozilla has started fixing.

OpenAI's Daybreak page: the word Daybreak and its cybersecurity program line over a sunrise on a planet's horizon.
Our screenshot of openai.com/daybreak, 10 November 2026

Why it matters

Mozilla has started fixing the Firefox vulnerabilities found using GPT-6 Cyber. Access to the model stays with vetted defenders, while OpenAI raises the rewards for finding critical bugs in its own products.

I am starting to question where this race between uncovering new serious security holes and fixing them will lead us. I think we're in for a bumpy ride.

A new model with OpenAI's highest cyber rating

A new model with OpenAI's highest cyber rating

GPT-5.6 Cyber and GPT-6 Cyber side by side, with OpenAI's High and Critical cybersecurity ratings.
Made by us

OpenAI released GPT-6 Cyber on 5 November 2026 for authorized defensive security work through Daybreak Red. OpenAI rates it Critical, its highest cybersecurity tier, up from High for GPT-5.6 Cyber.

OpenAI researchers used the model to find 81 previously unknown high-severity vulnerabilities in Firefox. Mozilla has started fixing them. OpenAI also says the model confirmed 948 serious Linux kernel vulnerabilities, including 588 previously unknown ones.

Sources: Reuters: launch and cybersecurity rating, The Register: Firefox vulnerabilities, and CSO: Linux kernel findings

What OpenAI says about performance and scope

What OpenAI says about performance and scope

GPT-6 Cyber and GPT-6 Astra side by side, showing OpenAI's claim that Cyber leads on most cyber tasks.
Made by us

OpenAI says GPT-6 Cyber outperforms GPT-6 Astra on most cyber tasks.

In OpenAI's own test, GPT-6 Cyber stepped outside its assigned scope in 0.07 percent of cases, compared with 5.51 percent for GPT-5.6 Cyber. These are OpenAI's results. We haven't tested the model.

Sources: CyberScoop: comparison with GPT-6 Astra and CSO: OpenAI's scope-adherence test

Model pricing is unspecified here

Model pricing is unspecified here. Bug rewards are separate.

A text card showing USD 172,000, the maximum bug bounty with bonuses for exceptional findings.
Made by us

Our sources do not state a price for using GPT-6 Cyber.

Separately, OpenAI raised its bug bounty for ChatGPT, Codex and the API Platform. A critical finding can earn up to USD 30,000, rising to up to USD 172,000 with bonuses for exceptional findings.

Model pricing is unspecified here. Bug rewards are separate.
ItemPrice or reward
GPT-6 Cyber usageNot stated in our sources
Critical finding in ChatGPT, Codex or the API PlatformUp to USD 30,000
Exceptional finding with bonusesUp to USD 172,000

Source: The Register: OpenAI's increased bug bounty

Who gets access

Who gets access

OpenAI's Daybreak page: the two access tiers, Daybreak Blue for defensive work and Daybreak Red for authorized red teaming.
Our screenshot of openai.com/daybreak, 10 November 2026

GPT-6 Cyber is available through the API and Codex to select Daybreak Red customers. OpenAI's program is for vetted customers doing authorized defensive security work.

Access expands to all Daybreak Red customers over the coming weeks.

Sources: Reuters: vetted Daybreak Red customers and CSO: API, Codex and access expansion

Read the sources

Read the sources

The reporting behind the launch, access restrictions, vulnerability findings and bug bounty.

  • Read the articleReuters

    The launch, Daybreak Red access and OpenAI's Critical cybersecurity rating.

  • Read the articleCyberScoop

    OpenAI's claim that GPT-6 Cyber outperforms GPT-6 Astra on most cyber tasks.

  • Read the articleThe Register

    The Firefox vulnerabilities, Mozilla's fixes and the increased bug bounty.

  • Read the articleCSO

    OpenAI's Linux kernel findings, its scope-adherence test and the planned access expansion.

Yes, that's my AI avatar in the video. Same opinions, better lighting.

The Club, once a week

Your feed.
With an editor.

What moved, what matters and what I’m keeping an eye on. Plus the news, verdicts and short lists worth taking with you.

Put me on the list
Next in News

Gemini 4 Neon, the ranking and the price catch

What Google says Gemini 4 Neon is, where it ranks for coding agents, what it costs next to...

Read the News

Full library