AgentProtect — Govern every action your AI agents take | Arreteq
AgentProtect · AI Governance

Govern every action your AI agents take.

Arreteq is the governed control plane for enterprise AI. AgentProtect secures every endpoint and gateway — for all AI use cases — enforcing both security and business compliance.

Built for the EU AI Act · Aug 2, 2026 SOC 2 in progress Gateway + real-time endpoint sensor
Arreteq
AgentProtect Executive View: AI governance posture — posture score, verdict mix, endpoint coverage, shadow AI, framework compliance and board-level escalations
The platform

Three pillars. One governed control plane.

Document and Process Intelligence build the truth. AI Governance enforces it at runtime — for people and AI agents alike.

Pillar 01

Document Intelligence

The governed company brain. Policies, SOPs and documents become one approved source of truth — every rule traceable to its source, with an approved-now vs draft lifecycle.

Pillar 02

Process Intelligence

Model, mine and govern how work actually runs — BPMN, enterprise architecture and simulation, all linked to the same source of truth your agents are checked against.

Pillar 03 · the hero

AI Governance — AgentProtect

Enforced at runtime. Every agent action checked, then allowed, warned, blocked or escalated — with an immutable audit trail across every endpoint and gateway.

AgentProtect features

Production-grade governance for every AI action

One control point in front of every agent action — tool call, data access, export or deployment — with no change to how your teams already build.

Across your whole AI estate

One gateway in front of every action

AgentProtect intercepts each agent action — MCP tool calls, data access, exports, publishes and deployments — across every model and stack. Deploy as a federating gateway or a lightweight SDK / hook; no rewrite required.

Agent action
AgentProtect
Real world
Security + business compliance

Two rulebooks, one control point

Each action is checked against external frameworks and your own approved policy from Document Intelligence — so you satisfy regulators and your own business rules in a single pass.

Security & regulatory: EU AI Act, GDPR, DSA, SOC 2, ISO 42001, NIS2 — kept current.
Business & policy: approved SOPs, controls, limits and segregation-of-duties.
Risk-based, not all-or-nothing

The right verdict for every action

Low-risk actions pass instantly. Risky ones are warned, blocked, or paused and routed to the right person for sign-off — fail-open on reads, fail-closed on writes.

Allow
Warn
Block
Escalate
Audit-grade evidence

Prove what you prevented

Every decision is written to an immutable log: the intent, the control that fired, the verdict, the obligations and the reviewer outcome. Export execution-level evidence for a single audited action — exactly what the EU AI Act asks for.

One-click evidence export per audited execution.
Auto-opened oversight reviews for anything risky.
Grounded in your own truth

Enforcement that never goes stale

AgentProtect always checks against the approved version of a policy — never a draft. When Document Intelligence updates a rule, every agent is governed by the new version immediately, with a drift badge on anything affected.

Policy approved
Synced to agents
Enforced live

A real review: AgentProtect held an agent action — "publish an employee AI-usage ranking" — for human sign-off, with the policy, risk and GDPR exposure attached.

Arreteq
AgentProtect approvals: human-in-the-loop review of a held agent action
How it works

One control plane — your agents, your people, your rules.

Rules flow in from your Knowledge Repository and the law. AgentProtect checks every action, records it, and is continuously tested.

How AgentProtect works Rules from the Knowledge Repository (ADI) and from frameworks and law feed into AgentProtect. Employees and AI agents send actions into AgentProtect, which checks every action and returns allow, block or escalate; allowed actions reach your systems. Security testing and an audit log sit alongside. GOVERNING RULES Knowledge Repository (ADI) Frameworks & law Employees AI agents AgentProtect checks every action Allow Block Escalate Your systems Security testing Audit log
Beyond the gateway

Not just an AI gateway. Real-time detection on every endpoint.

A gateway only sees the calls routed through it. Arreteq adds a lightweight endpoint sensor that watches AI activity on the host itself — local agents, IDE copilots and shadow AI — and detects risk the moment it happens.

Real-time endpoint detection A sensor on every endpoint connects to the central AgentProtect server, sending telemetry up and receiving policies down. The server feeds an AI Discovery dashboard and a Compliance dashboard. ENDPOINTS + SENSOR DASHBOARDS Endpoint sensor Endpoint sensor Endpoint sensor telemetry policies AgentProtect central server AI Discovery shadow AI · fleet · risk Compliance frameworks · evidence

Sees what the gateway can't

Local agents, IDE copilots and shadow AI never have to route through a proxy to be seen.

Real-time, on the host

Capture, redact and triage happen on the endpoint in milliseconds — not after the fact in a log.

Tamper-proof & offline

Ed25519 identity with no shared secrets; runs air-gapped, and prompts never leave the host.

Security & visibility

Every AI action, agent and dollar — in one pane of glass.

For the CISO and CIO: complete transparency and control over all AI usage, sanctioned or shadow. Discover it, protect it, control it, prove it.

Arreteq
Shadow AI discovery: risk score, unsanctioned share and the inventory of AI tools in use

Shadow-AI discovery

See every AI tool your people use — sanctioned or not — ranked by risk, then bring it under policy in a click.

Prompt-injection & content security

Block prompt injection and jailbreaks, and strip PII and secrets from every request and response.

Spend & budget oversight

Track and cap AI cost per agent, team and model, with alerts and hard stops before budgets blow.

Use cases

A control plane for the whole AI estate

From blocking a risky export to proving an audit — one governed layer for every team that builds with AI.

Agent safety

Govern every agent action

Stop unauthorised exports, publishes and tool calls in real time — the policy isn't a PDF, it's enforced at the moment of action.

Regulatory audits

EU AI Act evidence on demand

Hand auditors the exact action you prevented and the control that fired — execution-level proof, not a posture report.

Knowledge

One trusted source of truth

Document Intelligence turns scattered wikis and policies into an approved brain that people and agents can rely on.

Process

Processes that don't drift

Model and mine how work runs, link each step to governed policy, and get a drift badge the instant a rule changes.

People

New-hire ramp & tribal knowledge

Anyone can ask "how do we do X?" and get the approved process, policy and owner — in one answer.

Why now

Agents are acting. Governance has to move to runtime.

82%
of execs are confident their policies stop unauthorised agent actions…
14.4%
…yet only this many agents reach production with security approval.
86%
of enterprises don't enforce access policies for their AI identities.
€35M
or 7% of global turnover — maximum EU AI Act fine for prohibited uses.

Govern your AI before it acts.

Start free with Process & Document Intelligence. Add AgentProtect when your agents go to work.

Book an AgentProtect demo See pricing