OpenAI pipeline patches 948 Linux kernel bugs for $1.2M

The Linux kernel run points to a different use of AI models than the open-weight releases and product launches that dominate the same news cycle. OpenAI frames the work as vulnerability discovery, exploitation and patching handled by an autonomous pipeline, and it says the agents found and used exploitation techniques that its team and consulted experts consider novel. That claim matters because it suggests the model is not simply replaying known attack patterns.

OpenAI also says it is working with Linux maintainers and Patch the Planet partners on reports and patches, which places the effort inside the usual disclosure process rather than around it. The company says the run used an unreleased model, separate from the rumored GPT‑6.2 Sol and Astra models, and that it is directing a significant amount of frontier compute toward pro bono open-source patching. Those details suggest the pipeline is still a research effort rather than a shipping product, though the cost figure gives a concrete sense of what a single large-scale run required.