Our response to the Axios developer tool compromise

OpenAI reported that a third‑party developer tool, Axios, was compromised as part of a broader software supply‑chain attack on March 31, 2026. The attack involved a GitHub Actions workflow used in the macOS app‑signing process that downloaded and executed a malicious version of Axios (v1.14.1). The workflow had access to the certificate and notarization material used to sign OpenAI macOS applications, including ChatGPT Desktop, Codex, Codex‑cli, and Atlas. OpenAI’s analysis concluded that the signing certificate was likely not exfiltrated due to timing and sequencing factors, but the company treated it as compromised and revoked and rotated it. Effective May 8, 2026, older macOS desktop apps signed with the old certificate will no longer receive updates or support and may become non‑functional. OpenAI published new builds of all relevant macOS products signed with the new certificate and worked with Apple to block further notarization using the impacted material. The root cause was a misconfiguration in the GitHub Actions workflow that used a floating tag instead of a specific commit hash and lacked a minimumReleaseAge setting. OpenAI confirmed no evidence of user data compromise, no unauthorized software modifications, and that the incident only affected macOS apps, not web versions. Users are advised to download updates from official sources and to update within a 30‑day window to minimize disruption.