OpenAI said Oct. 26 that its Codex Ultrafast mode cut some engineers’ internal security investigations from one or two hours to 10 to 15 minutes, according to the company’s blog post1.
The reported time reduction matters because OpenAI says engineers use the mode to investigate several possible causes at once when an incident unfolds. The company says that faster searches let staff stay focused on the problem instead of waiting for results or splitting attention between Codex and their own investigation.
OpenAI presented the account as part of a series of illustrative examples of employees using its own technology and APIs internally. The post describes Ultrafast as a high-speed mode in Codex and says it has become the default for incident response among OpenAI engineers. The example is about the company’s own operations, not a reported customer deployment.
AJ Trbojevic, whom the post identifies as a member of the technical staff and a datacenter security architect, said security investigations previously took an hour or two to gather and normalize structured and unstructured data, then provide context for a decision. Trbojevic said that process now takes 10 to 15 minutes and can at times happen “near real time.” The post does not provide independent performance tests.
Other employees described using Codex to search incident logs and Slack messages for possible causes, including references to changes or similar symptoms. Aaron Friel said he directs Codex to monitor an incident-response Slack channel and investigate questions as they arise. Another employee, Jeff Niu, said most of his work with Ultrafast now takes less than 30 seconds, a threshold at which he otherwise would check Slack or ask a colleague. The post also describes uses for planning, browser tasks and code refactoring, but gives no measured time savings for those activities.