“OpenAI has a more rigorous security posture for its coding agents than most other companies, mandating sandboxing and a formal safety review process for every model release to mitigate risks like prompt injection.”