Incident comms getting applause
Two recent engineering leaders publicly handled incidents with blunt transparency — OpenAI’s Codex lead reset rate limits admitting uncertainty, and Anthropic’s Claude Code lead emphasized process and automation after a deploy mistake — both posts were praised as examples of strong incident leadership. These moments are shaping norms for candid, corrective communications in model/dev teams. (x.com) (x.com)
Anthropic’s Claude Code release accidentally published a 59.8 MB source map that contained roughly 512,000 lines of TypeScript across about 1,900 files, exposing internal application code in the v2.1.88 package on March 31–April 1, 2026. (2oceansvibe.com) Boris Cherny, head/creator of Claude Code, publicly said the root cause was a manual deploy step that “should have been better automated” and framed the mistake as a systems/process issue rather than an individual error. (itpro.com) Cherny’s thread listed concrete next steps — adding integrity checks, removing the offending package, and using Claude Code itself to validate future releases — and Anthropic described the exposure as a packaging issue with no customer credentials disclosed. (news.aibase.com) Anthropic’s remediation included mass DMCA takedowns that temporarily disabled roughly 8,100 GitHub repositories before the company narrowed and retracted most notices, leaving enforcement targeted at a single repo and 96 forks. (techcrunch.com) OpenAI Codex’s engineering lead Thibault Sottiaux (Codex/MTS) confirmed a recent session-hang incident had stabilized and announced a planned reset of usage/rate limits tied to a plugin rollout on March 26–27, 2026, with community reports noting additional resets around April 1, 2026. (softwareengineeringdaily.com) Both public threads illustrate a repeatable incident-comms pattern: named engineering lead + rapid factual timeline, an explicit mitigation action (rate-limit reset or package removal), blameless framing that points to process rather than person, and a list of follow-up engineering controls (automation, integrity checks) — each element visible in Cherny’s and Sottiaux’s posts and the companies’ subsequent actions. (itpro.com)
