Users report Codex and Claude refusing reverse-engineering and malware-removal tasks; an OpenAI employee says it is accidental

Users on X and Reddit report that OpenAI's Codex and Anthropic's Claude Opus 5.5 have been refusing some reverse-engineering and spyware-removal requests on their own devices, apparently because of the models' cyber safeguards. These are user reports; neither vendor has announced a policy change.

Open-source developer Eric S. Raymond, in a post that drew about 1,700 likes by October 25, said he had heard both companies had "nerfed" their models so they block people cleaning spyware and spamware off their televisions, and urged them to stop. Other users posted examples: one said Opus 5.5 refused a JTAG task and flagged their own documents, another that cyber-enabled models refused to read logged malicious requests.

OpenAI employee roon replied on X that the blocks are "mostly accidental and a result of difficult decision boundaries", that there is no prohibition on reverse engineering with Codex, and that it is "getting fixed". It was an informal post, not an official OpenAI statement. Anthropic has not commented in the posts available.

Commentator Andrew Curran wrote that the Claude complaints began on Reddit, where many commenters say Claude will still do the task if the request is rephrased, and that he expects Anthropic to treat it the same way. That is his expectation, not something Anthropic has said.

On October 31, Tom's Hardware reported more user claims that Claude Code or Opus 5.5 refuses some game work. It cites a rule circulating on X, attributed to a user named Chasm, that tells the model to build original games but "not a recreation of a specific existing game"; the rule's text is unconfirmed. It also cites a number of refusal complaints on Claude Code's GitHub, including from developers working on anti-cheat and security reviews. Tom's Hardware says Anthropic has made no statement.