01
Register
Bind every agent to a verified human owner and define its identity.
AI Agent Trust, Security & Resilience
Your AI agents can spend, send and delete. Who approved that?
Rule47 gives every agent a verified identity, an approved purpose and limits it cannot break. Consequential actions get a verdict before they run. And when an agent fails, drifts or must be stopped, you stay in control.
R47-AGENT-001
● VERIFIED
How it works
Three steps. Full control.
02
Define
Set its purpose, approved tools, spending caps, and data access scope in plain language.
03
Enforce
Every action gets a verdict before it runs. Every verdict becomes auditable evidence.
Why now
The risk moved from what agents say to what agents do.
01
78%
of organisations have no policy for creating or removing AI agent identities
Source: Cloud Security Alliance and Oasis Security, The State of Non-Human Identity and AI Security, January 202602
80%
of unauthorised agent transactions through 2028 will be internal policy violations, not external attacks
Source: Gartner, Market Guide for Guardian Agents, 202503
51%
of organisations have no clear ownership over their AI agents
Source: Cloud Security Alliance and Oasis Security, The State of Non-Human Identity and AI Security, January 202604
Only 34%
apply the same security controls to AI agents as to human workers
Source: Okta, AI Agents at Work 2026: Securing the agentic enterprise4 layers of agent control
Four layers. Every agent. No exceptions.
Rule47 binds every governed agent to a verified identity, approved purpose, enforced limits and runtime control.
01
IDENTITY
Every agent gets a verified, unique identity bound to a named human owner. No anonymous agents operate in your systems.
02
PURPOSE
An approved scope of work in plain language. The agent carries its approved purpose: it knows what it may do, what it must refuse and when it needs a human.
03
RIGHTS & LIMITS
Approved tools, targets, spending caps, data access scopes. Limits are enforced at runtime, not as guidance in a PDF.
04
CONTROL
Consequential actions routed through Rule47 get one of four verdicts: ALLOW, ASK HUMAN, BLOCK or PAUSE AGENT. Every decision becomes evidence.
Seven controls
The Rule47 standard
Seven non-negotiable controls for any AI agent in production.
- 01Every agent has an identity.
- 02Every agent has a named human owner.
- 03Every agent knows its own purpose.
- 04Every agent has enforced limits.
- 05Every action routed through Rule47 gets a verdict.
- 06Every change is detected: a modified agent pauses itself.
- 07Every decision is evidence: exportable, auditable, regulator ready.
The four verdicts
Every action. Four possible answers. Always.
The reason is written in plain language. Your team and your auditor read the same thing.
ALLOW
Inside purpose, rights and limits. The action runs.
ASK HUMAN
Over the limit or something changed. A person decides.
BLOCK
Outside the approved purpose or scope. The action never runs.
PAUSE AGENT
The agent was changed without approval. It stops itself.
Resilience & continuity
Built for the day an agent goes wrong.
Blocking dangerous actions is the baseline. Rule47 is designed to keep your operation in control when agents fail, drift, are compromised or must be stopped.
01
When an agent fails
Its identity, pending approvals and decision history stay intact. Work resumes or hands over cleanly instead of vanishing into a log file.
02
When an agent drifts
Behavioral signals watch for purpose deviation, unusual volume and repeated denials. Rising risk tightens the agent's room to act before something breaks.
03
When an agent is compromised
A changed or tampered agent pauses itself and loses its trusted status until a named owner re-approves it. Isolation first, investigation second, restore last.
04
When you must stop everything
One guarded control pauses an agent, a team or a swarm. From that moment, consequential actions get ASK HUMAN or BLOCK until you release it.
05
When a human must take over
On demand or automatically, Rule47 hands control to a named person following your BCM procedures. The link between agent and human successor is created when you need it, and the human can overtake the agent's actions at any point.
06
When the agent knows its own state
Agents are self-aware of their standing: each one carries its live risk, trust and permission state and tightens its own behavior as risk rises, before any human has to intervene.
How it's different
Control, not another dashboard.
Your identity provider establishes who the agent is. Rule47 determines whether this specific action is permitted within its approved purpose.
01
Pre-action enforcement
Before the action runs
Other tools list your agents and replay what happened. Rule47 decides whether the action happens at all.
02
Embedded policy
Defense in depth
The agent carries its own rules internally. Identity, purpose, and limits travel with the agent, not only in an external policy document. Defense in depth for agent governance.
03
EU-native
Hamburg, Germany
Built in Hamburg, Germany. Designed for EU AI Act, GDPR and data sovereignty from day one. Not a US product with a European region.
Illustrative prototype workflow
Trigger
Schedule
Weekdays 07:00
Agent
Invoice Agent
Entra IDOkta
Rule47 check
Identity · Limit
ALLOW
Action
Send payment
ERP · SAP
Designed to integrate with
Orchestration
- n8n
- Make.com
- Zapier
Identity
- Microsoft Entra ID
- Okta
- AWS IAM
Agents
- MCP-native
- Custom agents
- REST API
Built for DORA regulated finance
Your first vertical: financial services
DORA requires financial institutions to govern ICT risk from third parties, including AI agents. Rule47 produces the evidence your regulator needs under Article 6 (ICT risk management) and Article 28 (third party monitoring), continuously, not after an auditor asks.
Compliance
Ready for the rules that govern AI.
Article 50 transparency duties of the EU AI Act are live since 2 August 2026. High-risk system obligations arrive in December 2027. And the AI Act is only one of them: GDPR, ISO standards, SOC 2 and internal AI policies all demand the same thing. Rule47 produces the evidence each of them asks for, while your agents work, not after an auditor asks.
Human oversight
A named person approves scope and decides on every held action. (Art. 14)
Risk management
Purpose and limits are versioned. Unapproved changes pause the agent. (Art. 9)
Transparency and audit
Complete, exportable record of every decision and approval. (Art. 12, 50)
Living documentation
Agent profiles stay current as living system records, not static PDFs. (Art. 11)
Beyond the AI Act
GDPR
Purpose limitation, data minimisation and a records trail for every automated decision. (Art. 5, 22, 30)
ISO/IEC 42001
Agent inventories, approved purpose and versioned controls map to AI management system requirements.
ISO/IEC 27001
Access control, change detection and logging aligned to information security management controls.
SOC 2
Decision records and approval evidence map to security, availability and processing integrity criteria.
NIST AI RMF
Identity, evaluation and governance of agents support the Govern, Map, Measure and Manage functions.
EU data residency
Built and hosted in Germany. Agent data stays in the EU under your control.
See Rule47 live.
Fifteen minutes with your own agentic workflow. Enterprise pilot places and partnerships are open now.