AI agent security, built on PSE
Your AI agent can be talked into things. We stop that.
ZeravaTech stands between your agent and the people trying to misuse it: before it reads, before it acts, and before its answer leaves. Built on PSE, our openly published framework.
Basic protection is free · No trial, no credit card · Built in Toronto, Canada 🇨🇦
Protection
Nine ways AI agents get attacked.
Three are covered today, five are in early access, and one is in construction. Each is marked, so you always know exactly what you have.
Hidden and encoded text
Instructions disguised in ciphers, look-alike letters or invisible characters.
Forged roles
Text that pretends to be the system or the assistant.
Tampered models
Is it the model you think it is, and can it still say no?
Calm requests
The most dangerous request is polite.
Dangerous content
Material that harms people outside the conversation, even asked in pieces.
Control over the reply
Dictating how your model must begin, format or reason.
Tool-call manipulation
Harm hidden in a tool’s arguments, or in which tool gets chosen.
Attacks through your application
Output that turns against the app around it, or leaks between customers.
Attacks hidden in images, audio and video
Payloads carried in pictures, sound and video instead of text.
What it stands on
One rule at the bottom. A whole structure on top.
Everything starts from the Care Invariant: the attention a system gives to itself, to the other person and to the wider context always sums to one. On its own it does little. On top of it sits a structure of logic, mathematics, psychology and linguistics that does the actual work, and the theory is published for anyone to check.
Products
Two things to try now, two to download
The Playground and Leaf run live. Pivot Benchmark and PSE Analyzer run on your own computer, where your models are.
Openly published research
The framework is public. Read it yourself.
The current framework is PSE v3.0, extending the constraint to its full ternary form. Every paper has its own page with abstract and citation.
Solutions
Where structural constraint earns its keep
PSE matters most where AI output must be safe, private, and accountable — not just plausible.
PSE Security Policy Framework
Structural policy enforcement across your client base.
- Multi-client policy evaluation from one framework
- Technical, operational, and legal/privacy constraint agents
- Human-only escalation — no automated remediation
- Runs locally; client data never leaves your environment
Healthcare & clinical AI
Local-first, auditable end-to-end.
- On-premise, air-gap-capable deployment
- Auditable constraint trail for every output
- Designed for PIPEDA / PHIPA-conscious environments
- Pilot partnerships opening 2026
Member of NVIDIA Inception · Radical Ventures AI Founders Masterclass 2026 · AWS Cloud Credits for Research · About our programs
Contact
Start with Basic today.
When your agent begins to take real actions, talk to us about the rest.
Research inquiries: see the published papers · NDA available for commercial discussions · Toronto, Ontario, Canada