The report
On November 4, OpenAI said approved Daybreak Red customers could use GPT-5.6 Cyber through Amazon Bedrock in AWS GovCloud (US-East) for authorized vulnerability research, exploit validation and mitigation. This is a specific deployment of the older cyber model. The November 5 GPT-6 Cyber launch describes initial API and Codex availability for selected researchers, not automatic GPT-6 Cyber access in GovCloud.
There are three different permissions in the announcement. OpenAI Daybreak approval establishes eligibility for a restricted model route. AWS account provisioning makes the Bedrock service available in that environment. Asset-specific authorization still determines whether a security test against a target is lawful and within the operator's remit. Passing the first two checks does not grant the third. A reproducible deployment description should name the model version, region and service, caller identity and target scope rather than saying simply that 'Daybreak is enabled'.
OpenAI says this Bedrock route does not use inference data for training and does not require customers to send the same inference data directly to OpenAI. That is a provider statement about the offering, not proof of a customer's chosen Bedrock logging, retention, access controls or every regulatory certification. The original announcement contains the stated limitations. The GPT-6 Cyber dossier separately analyzes exploit capability and scope behavior.