OpenAI has released GPT-5.6 Cyber, a model built for vulnerability research, penetration testing and incident response — but it is locked to a named list of consultancies and security vendors. UK infrastructure teams now need to work out whether their existing suppliers actually hold access, and what that changes in vulnerability management services contracts.
View the data behind this chart
| Consultancies/Integrators | Security Vendors | |
|---|---|---|
| Named approved partners | firms9 | firms7 |
What GPT-5.6 Cyber is, and why OpenAI gated it
GPT-5.6 Cyber is OpenAI's model for finding vulnerabilities, checking whether they can actually be exploited, identifying affected systems, drafting fixes and helping push those fixes into production. It is not a general ChatGPT feature. OpenAI has said explicitly that it will not hand the underlying model to regular users, citing security risks, and pointing to the reality that similar models have already been abused to help launch attacks.
Instead, OpenAI is routing access through named partners who build the model into their own security products, managed services and client engagements. For UK buyers, this means the model itself is invisible in a contract — what matters is whether the supplier delivering your pentest, red team exercise or incident response retainer is one of the approved names.
The approved list: consultancies, integrators and vendors
OpenAI's named consultancies and service integrators are Accenture, IBM, Capgemini, Cognizant, EY, KPMG, PwC, NCC Group and SpecterOps. On the vendor side, Palo Alto Networks, CrowdStrike, Cisco, Sophos, Akamai, Fortinet and Cloudflare are listed as supported security vendors rolling the model into their products.
If your organisation already works with any of these firms for testing, monitoring or response, the practical question is not whether AI capability exists somewhere in the stack — it is whether your specific engagement, contract tier or product SKU actually includes GPT-5.6 Cyber, and under what safeguards. Procurement teams should be asking this directly rather than assuming inclusion.
Daybreak Blue and Daybreak Red: matching tier to risk appetite
OpenAI delivers partner access through what it calls Daybreak Access, split into two tiers. Daybreak Blue covers a broad range of defensive security workloads, while Daybreak Red is reserved for more specialised, closely governed work. This tiering matters for UK buyers because it effectively sets a ceiling on what any given supplier can do with the model on your environment — a Daybreak Blue engagement is not the same risk profile as a Daybreak Red one, even if both sit under the same vendor name.
OpenAI has said safeguards may include identity verification, clearly defined testing scopes, logging, monitoring and human oversight, and that access to the underlying models stays with the approved partner rather than transferring to the customer. As OpenAI put it: "Access to the underlying models remains with the approved partner and is not transferred directly to the customer." Partners are expected to define engagement boundaries, review findings and apply their own expertise before any action is taken — which effectively keeps a human decision point between AI-generated findings and live remediation.
Aligning approval status with NCSC access-control expectations
This gatekeeping model maps directly onto principles UK organisations are already expected to follow. The NCSC Cyber Assessment Framework's identity and access control principle calls for least-privilege access, unique credentials, and removal of special privileges once they're no longer needed — the same governance logic OpenAI is applying at the model level by restricting GPT-5.6 Cyber to named partners rather than open distribution.
For organisations using external suppliers to run testing or incident response, the NCSC Cyber Resilience Audit scheme standard also expects clarity on scheme-partner scope. That means CISOs should be documenting, in writing, which named partner is delivering AI-assisted work, under which Daybreak tier, and how that maps to existing zero trust access policies rather than treating it as an informal tooling upgrade buried inside a managed service.
What changes in disclosure, pentesting and IR workflows
Because the model can determine whether a weakness is genuinely exploitable and identify which systems are affected, approved partners may move faster from discovery to validated finding than traditional manual testing allows. That has knock-on effects for vulnerability disclosure timelines, since a partner using GPT-5.6 Cyber could plausibly surface and confirm issues sooner — but your organisation's own patching and change-control cycle won't automatically speed up to match.
For incident response workflows, the emphasis on logging, monitoring and human oversight suggests engagements will come with more formal audit trails than before, which is useful evidence for regulators but also something contracts need to specify. Buyers relying on managed detection and response from a listed vendor should confirm whether AI-assisted validation is now part of the standard service or a chargeable add-on.
View the data behind this chart
| Layer | Detail |
|---|---|
| OpenAI GPT-5.6 Cyber model | Underlying model stays with OpenAI |
| Approved partner (Daybreak Blue/Red) | Consultancy or vendor holds access |
| Defined customer engagement | Scope, logging, human oversight applied |
An audit checklist for UK infrastructure teams
Before assuming any current supplier relationship includes GPT-5.6 Cyber, run a short internal audit. Check contracts against the named partner list, confirm which Daybreak tier applies to your engagement type, and ask suppliers to document the identity verification, scope and logging controls OpenAI describes. This is also a moment to revisit broader comprehensive cybersecurity solutions planning, since AI-assisted testing sits alongside wider shifts in AI-powered cyber defence strategies and growing concern over AI model security risks more broadly.
Organisations that don't currently use any of the named consultancies or vendors aren't locked out permanently — OpenAI says cybersecurity providers can apply to join the Daybreak Cyber Partner programme, and customers can access the technology through participating security providers. It's worth asking existing suppliers whether they intend to apply, since that changes the long-term shape of your vendor shortlist.
- 01BleepingComputer — OpenAI releases ChatGPT 5.6 Cyber, but it's only for approved users · 10 August 2026
- 02The Hacker News — OpenAI launches GPT-5.4-Cyber with Trusted Access for Cyber · 1 April 2026
- 03The Hacker News — OpenAI limits GPT-5.6 rollout as Sol arrives · 1 June 2026
- 04NCSC — CAF Objective B, Principle B2: Identity and Access Control · 1 January 2026
- 05NCSC — Cyber Resilience Audit Scheme Standard · 1 January 2026
- 06TechRadar — OpenAI unveils GPT-5.6 upgrades under government-approved access · 1 June 2026