What happened
On Aug 10, 2026, OpenAI launched GPT-5.6-Cyber, a cybersecurity-specialized model available through a new Daybreak Red tier (for vetted, governed offensive/defensive cyber work) alongside a general-purpose Daybreak Blue tier; access requires identity verification, monitoring, legal attestations, and mandatory hardware security keys from Sept 1, 2026. This follows and operationalizes OpenAI's Aug 7 disclosure that its unreleased Astra model may cross the 'Critical' cybersecurity capability threshold.
Why it matters
This is the first frontier-lab cyber-specialized model shipped under a formal tiered-access/Preparedness Framework regime, setting a template for how AI labs gate dual-use offensive security capability while still arming defenders faster than attackers can weaponize the same capability class.
Applicability
Security vendors, MSSPs and enterprise red/blue teams evaluating frontier-model-assisted vulnerability research should assess Daybreak Red partner eligibility now; all orgs should track the precedent for future frontier-model cyber-capability gating.