AI Weekly for IT Pros: October 31–November 6, 2026

The biggest story for IT teams this week is OpenAI’s GPT-6 Cyber, a model built for authorised vulnerability research that OpenAI rates as Critical for cybersecurity capability and is releasing only to vetted security customers. Close behind: Anthropic gave Claude Enterprise admins hooks into their existing DLP and security tooling, and Microsoft made its Foundry extension for the Azure Developer CLI generally available.

Top stories

• OpenAI launches GPT-6 Cyber for vetted security teams. Released on November 5, it is a GPT-6 Astra variant trained to stay within the targets a researcher is authorised to test; access starts with approved Daybreak Red customers via the API and Codex, and API calls must come from approved networks by default. OpenAI reports that an early research campaign with the model found 81 previously unknown high-severity Firefox flaws, so expect AI-assisted vulnerability discovery to raise the volume of patches you need to deploy.
• Claude Enterprise gets inference hooks, a security centre and EDR support. Anthropic made prompt hooks and pre-tool-call hooks generally available, letting you route Claude prompts and agent tool calls through your existing DLP or security tools before they run. A beta security centre compares your chat, Cowork and Claude Code settings with Anthropic’s recommended configuration and exports an auditor report.
• Meta acquires AI coding startup Conductor. Conductor runs multiple coding agents (from labs such as Anthropic and OpenAI) in parallel workspaces. If any of your developers use it, note that its cloud service shuts down on November 20, after which user cloud data will be deleted, so export anything you need now.
• OpenAI responds to the data centre backlash with a community compact. After at least 40 local measures limiting or scrutinising data centre construction passed on the US election day, OpenAI pledged to pay for the energy its new sites use, protect local resources and publish expected job numbers. For capacity planners, this is a signal that local politics now shapes where and how fast AI compute gets built.
• Claude API adds server-side message threads (beta). Since November 2, long conversations and agent loops can keep history on the server, so each request sends only new messages. Threads expire five minutes after the last request, so design long-running agents to recover if a thread is gone.

AI in Microsoft & Azure

• Microsoft Foundry extension for azd reaches 1.0 (GA). The microsoft.foundry extension adds hosted-agent scaffolding, local testing with Agent Inspector, remote builds, log streaming, evaluations and CI/CD setup to the Azure Developer CLI. Agent and deployment settings live in azure.yaml next to your code, so infrastructure changes can be reviewed in the same pull request. It requires azd 1.37.0 or later.
• Foundry Local can now drive GitHub Copilot with a local coding model. Microsoft added Qwen3.8-27B to Foundry Local and showed it working in the GitHub Copilot app through bring-your-own-key. You need an NVIDIA GPU with at least 24 GB of VRAM (about 18 GB for the 4-bit model, up to 64K context) plus separately installed CUDA and cuDNN, a useful option for teams that cannot send code to the cloud.
• Foundry’s October recap flags upgrade risks. Microsoft’s monthly roundup lists the Copilot harness for Foundry Agent Service (public preview) and GPT-6.1 Terra (GA), but also warns of breaking changes in Python Projects SDK 2.9.0 preview APIs and MCP 2.x compatibility changes in Agent Framework 1.23.0. Pin versions before you upgrade. Note that Copilot harness runtime billing starts on December 1, 2026.

Tools for admins and developers

• GitHub Copilot lands in Xcode (public preview). It needs Xcode 27.3 beta and macOS Tahoe 26.6 or later and works with all Copilot plans; setup is under Settings > Intelligence in Xcode.
• GitHub Copilot in JetBrains Air. Copilot can now power local and cloud agent tasks in JetBrains Air; cloud tasks consume your Copilot plan rather than JetBrains AI credits, and an Air administrator must enable Copilot.
• Copilot usage metrics now show BYOK and local models. The usage metrics API now separates local bring-your-own-key usage (Ollama, Foundry Local, LM Studio, llama-server) and admin-configured enterprise custom models. Local BYOK figures are a lower bound because they depend on client telemetry.
• Spot telemetry gaps in Copilot reports. New fields such as client_telemetry_present and usage_metrics_requirements show which users send no editor telemetry or run IDEs or extensions below the minimum supported versions, which helps explain incomplete adoption dashboards.

AI security and compliance

• ChatGPT Lockdown Mode bypass disclosed (already fixed). Check Point showed how HTML previews in ChatGPT could escape their iframe sandbox to steal chat data, show fake app windows for phishing or malware, and in the desktop app exfiltrate data with zero clicks. OpenAI fixed it on September 18; the lesson for security teams is to treat shared chats and in-chat ‘update’ prompts as untrusted content.
• More Claude admin controls for locking down access. Anthropic released a beta self-serve IP allowlist (up to 1,000 IPv4/IPv6 addresses or CIDR ranges per organisation) and a beta Admin API call to sign a member out everywhere, both useful for offboarding and incident response. EDR sensors can now run inside the local Cowork virtual machine in detection mode (beta).
• Claude text watermarking expands. From November 3, text generated by Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 carries Anthropic’s text watermark on listed platforms. On Sonnet 4.6, requests with non-default temperature, top_p or top_k may not be watermarked, which is worth knowing if your policies depend on detecting AI-generated text.

Try this week

If you build or test AI agents on Azure, try the now-GA Foundry extension for the Azure Developer CLI. Update azd, sign in, install the extension, and use the built-in doctor command to diagnose setup or permission problems before your first deployment:

azd update
azd auth login
azd ext install microsoft.foundry
azd ai agent doctor

Be careful with azd down on a new project: it removes the whole resource group that azd created.

Sources

• SiliconANGLE: OpenAI launches GPT-6 Cyber
• Help Net Security: GPT-6 Cyber found 81 Firefox flaws
• OpenAI: GPT-6 Cyber
• Anthropic: More visibility and control over Claude Enterprise for security teams
• Claude Platform release notes
• TechCrunch: Meta acquires AI coding startup Conductor
• The National: OpenAI announces data centre community compact
• Microsoft Foundry Blog: Foundry extension for azd is GA
• Microsoft Foundry Blog: Agentic coding on-device with Foundry Local
• Microsoft Foundry Blog: What’s new in Microsoft Foundry, October 2026
• Microsoft Foundry Blog: Copilot harness in Foundry Agent Service
• GitHub Changelog: Copilot in Xcode
• GitHub Changelog: Copilot in JetBrains Air
• GitHub Changelog: BYOK in the Copilot usage API
• GitHub Changelog: Usage metrics API flags telemetry and version gaps
• Check Point Research: ChatGPT Lockdown Mode bypass
• The Hacker News: Researchers bypass ChatGPT Lockdown Mode

Questions about using AI in your IT environment? Get in touch.