AI providers tighten limits on reverse engineering, sending some users to open weight models
The blocks may be partly accidental, but unpredictable guardrail changes are already pushing some practitioners toward local models.
AI providers have begun blocking a popular practice where users prompt Claude and OpenAI's Codex to reverse engineer hardware and software, such as unbricking smart TVs or removing ads from them. Users report safety checks in Claude Code and account restrictions for Codex, with at least one account banned for cyber abuse, and the Decomp.wtf project has shut down. An OpenAI engineer said the blocks are mostly accidental and that there is no prohibition on reverse engineering with Codex, adding that it is being fixed. Some practitioners are shifting to open weight, local models instead, citing unpredictable guardrail changes. Separately, Anthropic unveiled Claude Meetings, Sheets and Whiteboards aimed at non-technical knowledge workers, with Meetings limited to Teams and Enterprise plans and Sheets and Whiteboards in beta.
Why it matters
For developers who used frontier models to analyse devices they own, the rules now feel unreliable: a task that worked one week can be blocked the next, and an account can be restricted over cyber-abuse rules. Some are responding by moving to open weight models they can run themselves, where the terms do not change without notice. The episode shows how much of this work depends on provider policy that users cannot predict, and how quickly that policy can shift.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.