AI model
OpenAI model released to vetted partners only on the argument that offensive-security capability needs gating.
OpenAI will pair GPT-6 Cyber, its fourth security model of 2026, with its first product for deploying it, Fortune reports. With $1 billion pledged to subsidize critical-services users, OpenAI is moving toward selling security work to operators and gaining a view of how they use the model.
Reality
A 20% input and 33% output cut on GPT-5.6 Sol comes with a November 21 floor, while per-request regional routing makes data residency cheaper than plain global processing was in July.
Reality
The three escapes used a kernel bug disclosed weeks earlier, a libslirp flaw Debian 12 has yet to fix, and 0-days the agent found itself after QEMU was rebuilt from upstream. It is one researcher's account.
Reality
The invitation-only service takes read access to authorized repositories and ranks findings by whether the route is deployed, busy and already covered by a WAF rule, then proposes a fix, pre-checks it, and hands you the approval.
Reality
The billion is credit against OpenAI's own products rather than cash, aimed at water, grid, local government and open-source defenders, and OpenAI wants it consumed within six months, which turns model-assisted triage into a procurement question.
Perspective Coverage
4 publishersReality
GitHub's volume counters and Chainguard's account of agent-written code both point at dependency review as the step nobody is doing. Mandiant estimates mean time-to-exploit at minus seven days in 2025.
Reality
Anthropic's June decision to keep Fable usage logs for 30 days has turned zero data retention into a procurement gate at Nvidia, Booz Allen and Palantir. The metadata channel stays open under the clause they are demanding.
Perspective Coverage
4 publishersReality
The early-access service reads live traffic to work out which findings matter, then produces a scoped firewall rule and a draft patch for review, so the mitigation is what actually ships while the code fix waits on engineers.
Reality
CrowdStrike will run OpenAI's cyber-tuned model inside its own harness while policing OpenAI's Codex agents at runtime, which leaves one vendor configuring what agents can reach and auditing what they did.
Reality
Trail of Bits gave a preview model one job: escape the sandbox on its author's Debian 12 host. The route worth studying needed only a published CVE and an unlabelled upstream fix commit.
Reality
Greg Brockman warns an open-weight release due at the end of August will worsen the threat landscape, while OpenAI's strongest cyber model sits behind identity checks and hardware keys.
Reality
Client fixes shipped in June and July. The exploit route only went public this month, and Zoom scores the bugs well below the firm that found them.
Reality