Three ways teams bring us in.
Each track begins with a written scope, a fixed deliverable, and a specific artifact the team can hand to an auditor, a customer, or a procurement reviewer. No retainers, no open-ended engagements.
Teams with a launch, a red flag, or a review deadline.
Pre-launch agentic workflow, decision-support system, or internal copilot. Or: the team has already observed a jailbreak, prompt injection, or unstable tool output and needs a deeper diagnosis than the ticket.
A structural review of the system, not the model.
We read your system prompts, tool descriptions, scaffolds, and configs against the failure-mode taxonomy documented in our research. We run controlled adversarial probes. We write down what the system actually does under pressure.
Written record.
- Audit memo with prioritized findings and recommended fixes.
- Failure-mode inventory mapped to the system’s prompts and tools.
- Runtime-control recommendations the team can hand to engineering.
Teams moving from experimentation into production.
The agent has tool access. Traffic is real. You need controls that survive an incident review, not a demo.
Input-side and execution-side defense-in-depth.
We design or pilot runtime defenses around your existing stack: prompt-injection sensing at the boundary, policy enforcement on process and network calls, anti-fabrication guards on tool output, and signed evidence of what the system did and why.
Works with LangChain, Microsoft Semantic Kernel, AutoGen, LlamaIndex, and custom agent frameworks. Python, JS, TS.
Control plane the team can operate.
- Runtime control design with threat model, placement, and failure modes.
- Integrated pilot of little-canary and suy-sideguy or the team’s existing stack.
- Signed-receipt evidence pipeline, offline-verifiable.
Teams approaching procurement, governance, or regulator review.
You need a realistic gap view — not a compliance-theater sprint against a generic checklist.
Evidence posture that survives cross-framework review.
We map your current technical documentation to Annex IV, cross-walk it to ISO 42001 and NIST AI RMF, and identify the artifacts that are missing, weak, or likely to fail external challenge. The same evidence bundle is defensible under EU, ISO, and US audit review.
● High-risk AI system obligations take effect 2 August 2026 (EU AI Act Annex III).
Gap view and remediation path.
- Annex IV-oriented gap analysis with evidence provenance.
- Cross-framework map (EU AI Act · ISO 42001 · NIST AI RMF · SOC 2 AI).
- Remediation roadmap sized for your 2 Aug 2026 obligations.
A short email with what’s in scope, what triggered the review, and what evidence exists today is enough to open the conversation.
Email Hermes →