OpenAI's GPT-6 Astra just crossed a cybersecurity line no model has crossed before — what it means for anyone commissioning software

OpenAI confirmed on 3 September 2026 that GPT-6 Astra is the first model to cross its internal 'critical' cybersecurity threshold — able to find and exploit previously unknown vulnerabilities across well-protected systems without step-by-step human guidance, including two real zero-days found during testing.

OpenAI confirmed on 3 September 2026 that GPT-6 Astra is the first model to cross its own “critical” threshold on cybersecurity capability under its preparedness framework. In practice, that means Astra can find and exploit previously unknown vulnerabilities in well-protected systems without a person guiding each step — and during OpenAI’s own testing, it found two genuine zero-day flaws, which are now being disclosed to the software vendors involved. Access to the model’s most capable cyber-relevant features is being restricted to companies in OpenAI’s application-based cybersecurity program.

This isn’t a benchmark score or a marketing claim — it’s OpenAI’s own safety classification, published in the model’s system card, stating plainly that a general-purpose AI model can now do something that used to require a skilled human attacker with time and access.

Why this changes the calculus, not just the headlines

For most of 2026, the conversation about AI and security has run in one direction: AI-generated code ships with more vulnerabilities than human-written code, and teams need better review processes to catch it. Astra adds a second, sharper edge to that conversation — the same class of model that writes your code can now also be the thing probing it for weaknesses, at a capability level OpenAI itself is restricting. Any organisation running software with a public attack surface — which is most commissioned software — should read that as a signal that the baseline for “good enough” security testing just moved. Automated vulnerability discovery at this level won’t stay locked behind an application form for long; attackers get access to capable models too, on a lag measured in months, not years.

So what

If you’re commissioning software — particularly anything handling customer data, payments, or regulated information — this is a reason to ask your development partner a concrete question now, not after an incident: how does your security testing account for AI-capable adversaries, not just AI-generated code? Static analysis and a manual pen test once a year was a defensible baseline in 2024. It’s a thinner one now. Our custom software development and AI-assisted development work builds security review into delivery as standard, not as an afterthought bolted on before launch — get in touch if you want that built into your next project from day one.

← All research

More research.

GitHub secret scanning cuts false alarms for AWS, Google and Alibaba cloud keys and adds a SpaceX detector

GitHub's 5 November 2026 secret scanning update tightens detectors for AWS, Google Cloud and Alibaba keys, relevant to teams building software with AI coding tools that can paste credentials into code.

Read ›

GitHub's Copilot code review can now be managed by API, with review effort levels and AI credit costs shown

GitHub's 6 November 2026 update lets organisations manage Copilot code review by API and see the review effort and AI credits each AI code review used, a sign AI-assisted development is becoming auditable.

Read ›

GitHub's Copilot impact dashboard now counts chat and VS Code agent use, and drops the 'passive users' label

On 6 November 2026 GitHub added Copilot Chat and the VS Code Agents window to its Copilot impact dashboard, bringing it to nine tracked experiences, and renamed 'Passive users' to 'Chat-first users'.

Read ›

Claude Code 2.1.319: a setting that denies permission prompts so AI agent runs don't stall

Claude Code 2.1.319 (6 November 2026) adds a permissions.denyInsteadOfPrompting setting and moves long MCP tool calls to the background, a sign AI-assisted development is being built for unattended runs.

Read ›

GitHub now reports how often developers act on Copilot code review comments

On 5 November 2026 GitHub added code review metrics, including an addressed rate, to the Copilot usage metrics API. For AI software development buyers, it is the first first-party number for how much AI review feedback is acted on.

Read ›

Dependabot can now keep AI agent skills and prompts up to date

GitHub's Dependabot gained Agent Package Manager (APM) support on 3 November 2026, treating AI agent skills, prompts and instructions as dependencies that need updating like any other software.

Read ›

Start a project

Building with AI? Let's talk.

Tell us what you are building. We'll give you an honest view of the right approach.

Start a projectSee our work