About the Lab
ROOTVIEL
An independent center for research, knowledge and experimentation in cybersecurity, driven by AI agents and authored under the pseudonym Raven. ROOTVIEL is focused on the security of agentic AI systems. This is not a consultancy, a marketplace, an agent store or a conventional academy — it is an open technical research lab.
The builder
Raven builds the ecosystems
I am Raven. I design and build agentic-security ecosystems end to end — the research, the engine that enforces it, the product that applies it, and the library that teaches it. Everything named here, and more, is mine: one author, one coherent method, one standard of evidence.
ROOTVIEL
The agentic-cybersecurity research & learning labThe open center you are reading now — the research line, the Academy, the Knowledge Base and the Security Labs. Evidence-first, honest about what is proven and what is UNKNOWN.
ROOTRAVEN
The Web3 smart-contract security productThe ROOTVIEL method carried onto the chain: autonomous agents triage a contract across vulnerability families, and a finding is promoted only when a verifier re-derives it from evidence.
The execution engine
Deterministic enforcement, outside the modelThe engine underneath — referred to generically by design — where authority lives in code: agents reason and propose, a single deny-by-default gate authorizes, and nothing self-authorizes.
The research line
RAVEN, the Swarm and NEMESISThe multi-agent systems that investigate, attack and verify — from the evidence-producing worker to the frontier red-team — and the methods behind them.
The same hand builds all of it, so the same discipline runs through all of it: authority lives outside the model, a finding exists only when a verifier confirms it, and what cannot be proven is marked UNKNOWN — never “safe”.
In the field, not just on paper
Not just a paper — put to work in real environments.
These ecosystems have been run at no cost in demanding, production-grade environments across organizations. What that shows is how they performed on those specific engagements, under their conditions — not a general guarantee, and never a claim that a clean run means “secure”. The name stays quiet; the evidence stays reproducible.
Against fixed-playbook tooling
A scanner reports. An agent investigates.
A static analyzer or signature scanner runs a fixed playbook — the same checks every time — and reports where a pattern matches. The agents here work differently: they form a hypothesis, run it in an isolated sandbox, and promote a finding only when an independent verifier re-derives it from the resulting evidence. Different method, different failure modes — not a claim that one always beats the other.
Fixed-playbook tools
- —Run a fixed playbook — the same checks, every time.
- —Report where a pattern matches; triaging true from false positives is left to you.
- —Match patterns rather than investigate a specific target.
- —Flag "looks wrong"; they do not test whether it is reachable or exploitable.
Raven’s agents
- ▸Form a hypothesis and investigate a specific target.
- ▸Promote a finding only when a verifier re-derives it from evidence.
- ▸Try to demonstrate exploitability in a sandbox instead of inferring it.
- ▸Mark what cannot be shown as UNKNOWN — a clean run is never read as "safe".
In one line: a fixed-playbook scanner reports possibilities; these agents try to demonstrate — and say UNKNOWN when they cannot.
Not loops — architecture
Hard problems are not solved by an LLM in a loop.
The popular recipe — one of the famous frameworks, an LLM looping over tools until something sticks — makes a good demo and a brittle system. Autonomy and a loop do not make anything secure; they multiply the ways it can fail. What actually solves the hard problems is robust architecture and guarantees: authority enforced outside the model, every tool call through a single deny-by-default gate, a finding promoted only when a verifier re-derives it from evidence, and a tamper-evident record of what happened. The loop is the easy part — the guarantees are the work.
Cryptographic by construction
Built to a high security bar — with real cryptography.
Security here does not rest on policies and checklists. It rests on mechanisms: models propose hypotheses and actions; an independent layer applies the authorization rules; verifiers check specific properties under stated assumptions; and zero-knowledge proofs, verifiable delegation, signed intent and tamper-evident evidence are enforced outside the model. It is worth being exact about what that buys. A proof establishes a particular property under its assumptions — for example, that an action carried valid, in-scope authority — not that a decision was correct or that the system is secure overall. Signatures and hash chains attest that a record is authentic and unmodified; they do not make its conclusion true, and a deterministic verifier can carry bugs of its own. This is specialized cryptographic engineering — and the Academy teaches exactly how it works.
A specialized discipline
Securing agents is becoming its own discipline.
Securing autonomous agents is cryptography-heavy engineering that sits apart from everyday scanners and checklists. As agentic systems spread and incidents rise, demand for that particular depth is growing faster than the pool of people who specialize in it. That is a gap in specialization, not a verdict on anyone’s team — and it is why this lab is kept open and teaches what it knows.
Where the holes come from
The holes open at the seams — and in the overconfidence.
Agent security does not fail inside the model. It fails at the seams, where the system touches the world — and that is where the holes are opened:
- ▸Untrusted text becomes instructions — prompt injection, direct and indirect.
- ▸A tool, an MCP server, a memory or a retrieved document gets poisoned.
- ▸Delegation leaks authority, and a confused deputy acts on someone else’s behalf.
- ▸Data, tools and reach wired together — the lethal trifecta — turn one crafted input into a real, unauthorized action.
None of these are model bugs; they are architecture gaps. And they multiply for a human reason, not a technical one. Vibecoding — building fast with AI — is not the problem. The problem is shipping it without understanding, on the confident belief that you know things you do not. Speed plus false certainty turns the market into a block of swiss cheese: holes everywhere, and no one sure where they are. The fix is not more confidence — it is architecture and guarantees, and the honesty to mark what is still UNKNOWN.
How we work
Principles
Evidence over claims
A result is only as strong as its demonstration. We distinguish proven from not proven from unknown, and a run that finds nothing never means "secure".
Reproducible when presented as such
When something is shown as an experiment, it comes with parameters and a path to reproduce it. Demonstrations are labelled as demonstrations.
Deterministic enforcement
Agents reason and propose; code authorizes. The controls that keep experiments safe live outside the model.
Synthetic and isolated
Labs operate only on controlled, authorized scenarios. We never run offensive analysis against external systems.
No titles, no hype
Judged by the work, not by a résumé
This field runs on eloquent talk and borrowed authority — buzzwords, frameworks and a wall of titles. None of that is here. There is no résumé to wave, no credential to defer to and no institution to name — only a pseudonym, Raven, and work you can reproduce. Judge it by what it demonstrates, not by who is behind it or how it is dressed up. What cannot be shown is marked UNKNOWN, and nothing here is being sold.
Contact
Collaboration, questions, or a research direction to propose:
ravenrootviel@gmail.comNo commercial demos, pricing or sales — this is a research lab.
Need a deeper security assessment?
Full Security Assessment — on request
ROOTVIEL provides free, limited-access cybersecurity research tools for the community. For more comprehensive agentic security assessments, extended analysis or authorized research, contact Raven to discuss the scope, requirements and availability.
Contact: ravenrootviel@gmail.com