We build the stop button for AI systems
Fermi Technologies is a deep tech company working on the infrastructure that makes AI and cryptographic systems accountable. Kill Switch is our answer to a question that keeps coming up from the teams shipping agents: what happens when it goes wrong at three in the morning, and how do you show what happened afterwards?
Why now
Most companies deploying AI today are building on top of someone else's model. They can change a prompt and they can turn off a feature, but they have no way to stop a single harmful reply, no way to close one conversation for good, and no record they could hand to a regulator that could not have been edited after the fact.
The EU AI Act asks deployers for exactly those two things: tamper-evident logs, and human oversight that can intervene. Enterprise security reviews ask the same questions in different words. Kill Switch exists because the gap between those requirements and what a typical AI product can demonstrate is very wide, and closing it should not take a team six months.
What we care about
A control you can actually reach
Plenty of safety tooling scores a message and leaves you to act on it. The point of this product is the part after the score: stopping the message, ending the conversation, or halting everything, from code or from a page a non-engineer can open.
Evidence over assurances
We would rather hand you a log you can verify than ask you to trust our dashboard. That is why update and delete are revoked on the event tables for every role, ours included, and why the nightly verification records its verdict whether it passes or fails.
Honest about what it is not
This is a safety control, not a guarantee. The scorer is a language model and it will be wrong sometimes, in both directions. We say so in the terms, in the docs, and here.
Your content, your infrastructure
The scorer is a configuration value, not a lock-in. Point it at our hosted model to start and at your own GPUs when it matters, without touching a line of your code.
The company
Fermi Technologies Inc. is a Delaware corporation. Alongside Kill Switch we work on post-quantum cryptography and verifiable ledgers, which is where the tamper-evident side of this product comes from: the same anchoring approach that proves a chain was not rewritten is what we intend to use to prove that we did not rewrite yours either.
If you are deploying agents and any of this sounds like a problem you have, we would like to hear from you. We work closely with early customers and the roadmap follows what they actually run into.