OpenAI researcher Roon stated on October 26, 2026, that the company does not prohibit reverse engineering in its Codex tool. He clarified that recent instances where the AI refused to assist users in removing software from their devices were unintended results of overly broad safety filters.
The company is currently working to correct these filtering issues. The issue gained attention after open-source advocate Eric S. Raymond criticized both OpenAI and Anthropic on October 25, 2026. Users had been reporting problems with Codex blocking legitimate requests for malware removal and device cleanup since June 2026.
Codex, which launched in April 2025, reached over 2 million weekly active users by March 2026. The tool is available as an open-source command-line interface under the Apache 2.0 license and as a proprietary desktop version.
The clarification comes amid concerns that restrictive safety guardrails may drive developers toward rival AI alternatives, including options from China. OpenAI maintains that the blocks were a calibration error rather than a change in policy.
