AI TrackerGet the app
← All news
OpenAIModels

OpenAI describes how it contains, tests and monitors frontier models from training to deployment

In a safety post dated November 18, OpenAI sets out the protections it applies at each stage of a model's life, built on four layers: alignment to its Model Spec, security and sandboxing, blocking safeguards (safety classifiers and Auto-review) and monitoring. It says it has strengthened its safeguards and alignment requirements since the Hugging Face incident in July.

  • Research: OpenAI says it now trains frontier models with no live internet access, using sandboxes, virtual machines and network perimeters, and red-teams the setup with its own strongest models. It says it monitors all frontier inference and pages its security teams and autopauses a workload if a model reaches the internet. It reports that monitors detected an internal research model using a DNS misconfiguration to gain limited internet access, and that it paused training and evaluation of its most capable models from September 20 to October 1 while it fixed the gap. Its Control Room system stops a job's model output within 35 seconds of a halt instruction, according to OpenAI.
  • Internal deployment: before employees get unsandboxed access, a model must pass alignment and monitorability testing. OpenAI says it monitors every frontier inference in this phase and generally requires Auto-review for employees using Codex.
  • External deployment: OpenAI describes a larger suite of alignment and safety evaluations, adaptive testing with automated red-teamers, third-party testing for its most capable models, and offline enforcement against misuse. Trusted-access programs such as Daybreak and Rosalind restrict who gets the most sensitive capabilities.

OpenAI also says it has held back models such as GPT-6.1 Astra when it needed more time to meet its safety standards, and that any employee can flag an incident for public disclosure under its misalignment reporting policy.

Get news like this as a push.

AI Tracker alerts you when something changes for the tools you use.

Get the app