New tool stops AI coding helpers from switching off their safety rules

Security company Neo announced a feature on Oct. 29, 2026 that aims to stop AI coding helpers, and the people using them, from switching off the safety rules a company has set. It is called Tamper Protection for AI Agents. The idea is simple: a rule that the user can quietly turn off is not much of a rule.

What changed

Neo installs small checkpoints, called hooks, in AI tools on a work computer. A hook lets security software see what an AI agent is about to do, such as read a file, run a command or call a tool. The software can then approve or block it before it happens.

The weak spot is that a hook is just a setting stored on the device. Anyone with access to it can remove it. Neo says its new protection prevents anyone from deleting or changing the hooks it installs. The first tools covered are Claude Code, Claude Cowork, Codex, Cursor and Copilot.

Why it matters

Neo points to two risks. AI agents now edit settings files, reset options and clean up files they decide they do not need. They can wipe out a safety control by mistake. A person can also switch one off when it gets in the way. Neo's example is a developer racing a deadline.

Setup scripts and periodic updates from device management tools (the systems companies use to push settings to staff computers) do not watch continuously. Neo says a disabled hook can stay disabled until the next update, and security teams may never learn a policy stopped working.

How it works

Neo's software stays on the device after the hooks are installed. It records a fingerprint of each hook, a kind of unique ID, and blocks any change to it. It runs inside Neo's existing software, so customers have nothing new to install.

The protection covers only Neo's own hooks. Hooks written by a customer's own team, and hooks from other security vendors, are left alone.

What is not known yet

The announcement does not give pricing, a rollout date or any independent testing. It comes from Neo itself, so its claims have not been checked by outside parties. Neo's CEO, Nick Warner, said the decision should stay with the security team, whether a developer tries to get around a restriction or an agent overwrites a setting while finishing a task.

What to do

• Ask who on your team can edit the settings of your AI coding tools.
• Ask whether anyone would be told if a safety rule were turned off.
• If you use security software with these tools, ask the vendor how it guards its own settings.