Tech trends, week of Nov 6, 2026: AI security research, audit skills, Images 2.6

Three developments from the past week stand out for engineers and learners. Two are about AI doing security work. One is about the image APIs you may already be building on. All data is as of November 6, 2026.

OpenAI's Linux kernel research: 948 root exploits

On November 4, OpenAI published results from a controlled research run against the Linux kernel. In mid-October, a prototype pipeline using an upcoming (unreleased) cyber model analyzed 2,308 candidate findings. It dynamically reproduced 1,724 vulnerabilities and produced 948 local privilege escalation exploits. Each exploit could give an unprivileged local user root access. The run cost about $1.2M in API credits, roughly $1,280 per vulnerability.

Three caveats keep this in perspective. OpenAI's own security researchers reviewed the exploits and fixes before anything went to kernel maintainers. The patches are candidates for maintainer review, not fixes that have already shipped. The model has no public name, price or release date.

For engineers, the lesson is about economics. Finding exploitable bugs at scale now costs far less than it did, so the bottleneck shifts to triage, patching and designing systems with a smaller attack surface. If you're learning, spend time on secure coding, memory-safety basics and how to verify automated findings.

Link: https://openai.com/index/a-thousand-fewer-paths-to-root/

Cloudflare's security-audit skill

This repo is not new. It was created in June 2026 alongside a Cloudflare post about its vulnerability harness. But Cloudflare's developer account pointed people to it on November 4, in the middle of a wave of vulnerability disclosures, so it belongs in this week's picture.

It's an MIT-licensed skill for coding agents. It guides an audit through six phases: map the codebase, hunt for vulnerabilities, challenge each candidate, record findings, verify them, then report what is confirmed, rejected or still needs validation. You install it with the Skills CLI. Cloudflare reports that a single run finds roughly half of what repeated runs find in total, so plan on several passes.

Two cautions. The skill is the single-repo ancestor of Cloudflare's larger internal harness, not the harness itself. Its output also depends on the underlying model, and I found no published false-positive rates. Use it on your own projects, check each finding by hand, and use the exercise to learn how adversarial validation works.

Link: https://github.com/cloudflare/security-audit-skill

OpenAI Images 2.6

Also on November 4, OpenAI released Images 2.6 in ChatGPT, Codex and the API. The headline is generation that is 50% faster than before, plus GIF creation. In the API there are two models. GPT-Image-2.6 Flare is the faster option. GPT-Image-2.6 Sunburst gives more detail and keeps subjects more consistent across edits when quality is set to medium or high.

Developer-facing additions include marked-up images that show exactly where an edit should happen, SVG export, and a complexity parameter that controls how much planning the model applies. OpenAI suggests low complexity for quick edits and medium or high for scenes with many constraints.

Reported pricing is $8 per million image-input tokens, $30 per million image-output tokens and $5 per million text-input tokens for both models. That figure comes from press coverage of OpenAI's pricing page, so check the page before budgeting. A reasonable approach is to prototype with Flare, then switch to Sunburst for workflows with many sequential edits, and measure cost per finished asset rather than per call.

Link: https://openai.com/index/introducing-images-2-6/

How we picked these

Criteria: announced or actively promoted between about October 30 and November 6, 2026; backed by an official vendor page or repository; and practical for working engineers or learners. I left out items featured recently and items that were funding news, opinion or promotion without a technical source. Figures such as exploit counts, run cost and model names come from OpenAI's and Cloudflare's own publications, with press coverage used only to cross-check. The Cloudflare repo's June creation date is noted above. Data is as of November 6, 2026, and details such as pricing can change.

Sources

• OpenAI, A thousand fewer paths to root (Nov 4, 2026): https://openai.com/index/a-thousand-fewer-paths-to-root/
• Cloudflare, security-audit-skill on GitHub: https://github.com/cloudflare/security-audit-skill
• Cloudflare Developers on X (Nov 4, 2026): https://x.com/CloudflareDev/status/2118013149538390253
• OpenAI, Introducing Images 2.6 (Nov 4, 2026): https://openai.com/index/introducing-images-2-6/
• OpenAI Developers on X (Nov 4, 2026): https://x.com/OpenAIDevs/status/2118124753340756055