'Reverse Engineer Anything': AI Companies Crack Down on the Coolest Thing People Were Doing With AI
Reported by 404 Media and aiweekly2 sources ↓
Users trying to reverse-engineer devices with Claude and Codex have reported safety blocks, warnings or account restrictions. An OpenAI engineer says Codex blocks are mostly accidental and are being fixed.
Key points
- Claude users report safety checks preventing AI-assisted reverse-engineering
- Codex users report warnings, account restrictions and a ban
- OpenAI engineer Tarun Gogineni says reverse-engineering is not prohibited in Codex
- Some practitioners are turning to open-weight models for device research
Sources
Details
Reports from Claude and Codex users describe blocks or account restrictions when they tried to reverse-engineer hardware or software. Examples include work on smart TVs, IoT devices and users’ own code. The project Decomp.wtf, which supported software and game reverse-engineering, has also shut down in recent days, 404 Media reports.
OpenAI engineer Tarun Gogineni said the Codex blocks were mostly accidental, caused by difficult decision boundaries, and that the company was fixing them. Matt Brown, a security consultant quoted by 404 Media, said he had encountered blocks associated with cyber safety checks and was using the open-weight model GLM 5.3 to find hidden Bluetooth commands in an IoT gas monitor.
Written from coverage as of