A Leeds cybersecurity company has launched a protocol intended to shut down a compromised artificial intelligence agent without pulling the plug on it, aimed at the problem legislators have been circling with proposals for mandatory kill switches, according to Cavero Secure.
An autonomous agent executing trades, managing infrastructure or altering databases cannot easily be switched off once it goes wrong. Hard termination in a cloud-native environment risks corrupting state data, severing dependent supply chains and setting off cascading failures, which makes the obvious remedy an expensive one.
The company’s approach treats identity as a continuous cryptographic history of an agent’s state transitions from the moment it was created, instead of a static secret such as an API key or session token. Trust is re-evaluated action by action, and keys and authority are derived dynamically across interactions, so there are no stored credentials for an attacker to harvest or clone.
An attacker running a stolen or copied agent cannot do so without diverging from the legitimate history, and the contradiction between the two timelines is detected before unauthorised actions clear. Cavero calls this the clone paradox.
Halting an agent is handled through what the company calls a history shift, in which modifying the local verification ledger leaves the target unable to prove its continuous historical state, at which point it fails its mathematical challenge and becomes invisible to the network, isolated without crashing host servers or affecting adjacent workloads. Because the suspension is anchored in state continuity instead of process destruction, it can be reversed, and an administrator can roll the agent back to its last known good state.
Professor Ben Varcoe, co-founder of Cavero Secure, said: “The industry has spent years trying to govern autonomous AI using 20th-century credential models and blunt emergency stop buttons. Security does not come from building higher walls or pulling physical plugs, it comes from continuous identity verification. By anchoring authority in an unbroken, living state history, CaveroCTX gives enterprises absolute control and surgical precision over their autonomous digital workforce.”
The protocol is being deployed with selected enterprise design partners in financial services, cloud infrastructure and defence technology. Cavero Secure describes itself as working on post-quantum key distribution and continuous trust protocols, out of advanced quantum information research.
