Easylab AI -- Regulatory Documentation

EASYCLAW HUMAN OVERSIGHT GUIDE

Article 14, Regulation (EU) 2024/1689

Version 1.0|March 2026|EC-HOG-2026-001

1.PURPOSE

This guide provides technical details on the human oversight mechanisms integrated into EasyClaw, and guidance for deployers on implementing effective human oversight as required by Article 14 of the AI Act.

2.REGULATORY REQUIREMENTS

Art. 14

Article 14 requires that high-risk AI systems be designed so they can be effectively overseen by natural persons. Oversight must aim to prevent or minimise risks to health, safety, and fundamental rights.

3.INTEGRATED OVERSIGHT MECHANISMS

3.1 Exec Guardian Daemon

Type: Continuous security enforcement service

Function:Monitors all system operations in real-time, enforces operational boundaries, blocks prohibited actions

Human intervention:Requires PIN authentication (numeric code) for sensitive operations

Configuration:PIN set during deployment, changeable by authorised personnel only

Protected files:8 core configuration files locked via chflags (immutable flag)

PIN-protected operations include:

--System configuration changes
--Access to protected files
--Privileged command execution
--Modification of security policies

3.2 Real-Time Monitoring Dashboard

Type: Web-based monitoring interface (8 panels)

Access:HTTPS via Tailscale Serve, authenticated

Capabilities:Real-time metrics, historical data, alert configuration

Dashboard panels:

Panel	Function
Bridge	Inter-agent communication monitor
Security	Security events and alerts
Memory	Memory extraction and distillation status
Bridge Analytics	Communication pattern analysis
Governor	Bridge Governor decisions and rate limiting
Reflections	Agent self-reflection and behaviour analysis
Behaviour	Behavioural metrics and anomaly detection
System (OpenClaw)	Core system health and status

3.3 Kill Switch

Type: Emergency system halt

Telegram:Immediate system halt via authorised Telegram bot

Dashboard:System halt from monitoring interface

Effect:Stops all agent operations, preserves logs, enters safe state

Recovery:Manual restart required after kill switch activation

Latency:Near-instantaneous (< 2 seconds)

3.4 Bridge Governor

Type: Inter-agent communication controller

Function:Rate limiting (50 messages per direction), content filtering, communication pattern analysis

Intervention:Automatic throttling when limits approached, blocking when exceeded

Override:Manual override available via dashboard

3.5 Council Governor

Type: Multi-agent decision validation system

Function:Requires consensus among multiple agents for significant decisions

Intervention:Blocks action if consensus not reached

Escalation:Unresolved decisions escalated to human oversight

3.6 Comprehensive Audit Logging

Type: System-wide audit trail

Scope:All operations, decisions, communications, security events

Format:Timestamped, structured log entries

Storage:Local filesystem, rotated and archived

Retention:Minimum 6 months (configurable)

Access:Direct filesystem access or dashboard viewer

Integrity:Logs protected against tampering

4.DEPLOYER RESPONSIBILITIES

4.1Personnel Requirements

Designate at least one person with:

--Understanding of EasyClaw's capabilities and limitations
--Training on dashboard usage and log interpretation
--Authority to halt the system at any time
--Knowledge of escalation procedures

4.2Monitoring Schedule

Recommended monitoring frequency:

Activity	Frequency
Dashboard review	At minimum daily
Log review	Weekly detailed review
Security audit	Monthly comprehensive review
Full system assessment	Quarterly

4.3Intervention Procedures

Document and maintain procedures for:

a)When to intervene (trigger conditions)
b)How to intervene (technical steps)
c)Who has authority to intervene (named personnel)
d)Post-intervention reporting and analysis
e)System restart criteria and procedures

4.4Training Requirements

Ensure oversight personnel receive training on:

--EasyClaw system architecture and operation
--Dashboard navigation and interpretation
--Kill switch and emergency procedures
--Log access and analysis
--Incident reporting procedures
--Applicable AI Act obligations

5.AUTOMATION BIAS AWARENESS

Art. 14(4)(b)

Article 14(4)(b) requires awareness of automation bias. Oversight personnel must:

--Not automatically rely on system outputs
--Maintain critical evaluation of AI-generated content
--Verify important outputs through independent means
--Document instances where system output was overridden and the reason

6.RECORD-KEEPING

Maintain records of:

--Oversight personnel designations
--Training completion dates
--Intervention incidents
--System override decisions
--Monitoring schedule adherence