Does OpenAI Codex (ChatGPT Business / Enterprise / Edu & API) train on your code?

Trains on code

No

Default

Not trained on by default; orgs are opted out of data sharing unless they explicitly opt in (e.g. Playground feedback).

How to opt out

Nothing to do. Don't opt into data sharing / feedback programs.

Retention

ChatGPT Business/Enterprise: admins control retention; deleted conversations removed within 30 days. API: inputs/outputs kept up to 30 days for abuse monitoring.

Telemetry

Codex CLI anonymous metrics by default.

Disable telemetry

[analytics] enabled = false in config.toml

Zero data retention

yes · API: Zero Data Retention / Modified Abuse Monitoring for approved orgs on eligible endpoints.

Offline / local model

yes · Codex CLI --oss with Ollama or LM Studio.

Self-hostable

partial · CLI local; Codex cloud tasks hosted.

Score

87/100 · training 40/40 · retention 17/25 · offline 20/20 · telemetry 10/15

Classes: training No · retention ≤ 30 days · telemetry On, can disable · offline yes

Sources