Security and trust
Responsible AI
Our products put AI into the engineering of aircraft, vehicles, medical devices, power systems and banks. These are the principles we build them by, and the product behaviour behind each one.
[Draft: to be approved by Neural Inverse before publication]
A person decides
AI proposes; people decide. An engineer confirms every write to hardware, and the agent actions your policy marks wait for a named approver. A recorded approval counts only for exactly the action it names, and the person who asked for a controlled change cannot approve it. Each agent has its own identity.
Every action on record
Every model request and agent action is attributed to a person or an agent identity and written to a tamper-evident audit log that can be verified offline. You can show an assessor who asked for a change, what the agent did, what checked it and who approved it.
Your models, your data
Engineers and agents reach models only through the gateway you control, from the list you approve. You can use your own provider agreements or models you host yourself, and run the whole platform on a network with no internet connection. Prompts, tool calls and context pass data loss prevention before they reach a model. We do not use your code, prompts or data to train AI models.
Your tools verify
AI output is not evidence by itself. Your qualified compilers, simulators and analysers run the checks, and their results are kept with the change. We do not replace the qualification of your tools or your processes.
Claims we can show
We describe only what works today, and say what is still in development. We produce evidence toward standards and never claim that our products, or your use of them, are certified; certification stays with you and your assessor. Each check and evidence record says what it looked at and what it does not prove.
Known limits
Models make mistakes, including confident ones. Agents can misread a requirement, and policy can only govern the actions it covers. Treat AI output as a proposal from a colleague whose work you review, keep approvals on the actions that matter, and keep your verification independent of the model.
Raising a concern
If you believe one of our products behaves in a way that is unsafe or contrary to these principles, email legal@neuralinverse.com. For security vulnerabilities, follow the vulnerability disclosure policy.