OpenAI's internal incident-response agent, built on Codex. When an incident is detected, it collects context, proposes mitigations, and answers questions in the incident Slack channel. The report describes an engineer instructing it to apply a chosen mitigation.
Proposes mitigations in the incident channel; an engineer can instruct it to apply a specific mitigation[Source 1]
Answer developers' questions
Answers developers' questions as part of the incident Slack channel[Source 1]
Where people stay involved
People review, approve, or guide the agent’s work here.
Reported: The engineer-controlled mitigation instruction is the documented checkpoint; the article states Sevbot never executes a mitigation on its own.
Not reported: The article documents no check of Sevbot's proposals or answers; the propose-only limit is a behavior boundary, not a validation mechanism.
Reported observations
Not reported: No outcome measurement for Sevbot appears in the reviewed captures. The certificate and commit records evidence a deployed service, not results.
Lessons
OpenAI describes autonomous mitigation of routine outages as a goal for Sevbot; on-call duty still exists in the reported deployment.[Source 1]
Reported: The article lists the run in order: wake, collect, propose, answer, and the engineer instruction. The propose-only primitive spans the proposal and the instruction the article states separately.
human involvement
Reported: The engineer-controlled mitigation instruction is the documented checkpoint; the article states Sevbot never executes a mitigation on its own.
implementation
Reported
validation
Not reported: The article documents no check of Sevbot's proposals or answers; the propose-only limit is a behavior boundary, not a validation mechanism.
observations
Not reported: No outcome measurement for Sevbot appears in the reviewed captures. The certificate and commit records evidence a deployed service, not results.
SupportsCertificate Transparency records for sevbot.api.openai.comPreserved content.md, lines 13-41 (certificate table for sevbot.api.openai.com; earliest not-before 2025-11-20 at lines 37-41, with renewals through 2026-08-01; evidences a deployed OpenAI service under the Sevbot name, not its incident-response behavior)
The source explicitly contrasts the hoped-for autonomous response with the continuing on-call requirement, so the goal must not read as current behavior.
The article says Sevbot proposes mitigations and that an engineer can tell it to apply one. This supports an engineer-controlled mitigation path; the source's 'never executes any' wording is read as a limit on autonomous execution.
Observation date
2026-09
SupportsInside OpenAI's agentic software factoryPreserved content.md, lines 190-193 (determines possible mitigations but never executes any; an engineer can tell it to apply a specific mitigation)
