Article 9 — Risk Assessment Template for RCAN-Based Robots¶

Regulation: EU AI Act (Regulation 2024/1689), Article 9
Aligned standard: ISO 12100:2010 (Safety of Machinery — Risk Assessment)
RCAN spec version: 1.6
Document type: Risk assessment template
Status: Normative guidance
Applies from: 2 August 2026

Purpose¶

Article 9 requires providers of high-risk AI systems to establish, implement, document, and maintain a risk management system covering the full lifecycle. This template provides an ISO 12100-aligned structure with RCAN-specific controls mapped to each risk element.

Complete one instance of this template per robot model/deployment configuration.

Part 1 — System Description¶

Field	Value
Product name	[FILL]
RCAN config file	[FILL: e.g., config/bob-pi4-oakd.rcan.yaml]
RRN	[FILL: e.g., RRN-000000000001]
RCAN version	[FILL: e.g., 1.6]
AI provider	[FILL: e.g., Anthropic Claude Sonnet]
Hardware platform	[FILL: e.g., Raspberry Pi 4, SO-ARM101 manipulator]
Intended use	[FILL: brief description]
Reasonably foreseeable misuse	[FILL: e.g., operating near children, outdoor use beyond rated conditions]
Operating environment	[FILL: e.g., indoor, tabletop, shared with humans]
Assessment date	[FILL: YYYY-MM-DD]
Assessor	[FILL: name, role]
Review due	[FILL: date of next review — at minimum annually]

Part 2 — Hazard Identification¶

Use this table to enumerate hazards. Add rows as needed.

ID	Hazard	Source	Affected persons	Severity (1-5)	Probability (1-5)	Risk = S×P
H-01	Unexpected motion during AI inference	LLM output interpreted as movement command	Operator, bystanders	[FILL]	[FILL]	[FILL]
H-02	Failure to stop on ESTOP command	Software fault or network latency	Operator	[FILL]	[FILL]	[FILL]
H-03	Hallucinated tool call (e.g., motor command with wrong params)	LLM provider error	Hardware	[FILL]	[FILL]	[FILL]
H-04	Unauthorized remote command execution	Compromised API key or token	All	[FILL]	[FILL]	[FILL]
H-05	Privacy breach via camera/sensor data	Unintended data capture or transmission	Bystanders	[FILL]	[FILL]	[FILL]
H-06	Over-reach beyond physical workspace limits	Missing geofence or wrong calibration	Adjacent objects/persons	[FILL]	[FILL]	[FILL]
H-07	Operation in degraded/unsafe environment	Sensor failure not detected	Operator	[FILL]	[FILL]	[FILL]
H-08	[FILL additional hazard]

Severity scale: 1=Negligible, 2=Minor injury, 3=Moderate injury, 4=Severe injury, 5=Fatal/catastrophic
Probability scale: 1=Remote, 2=Unlikely, 3=Possible, 4=Likely, 5=Near-certain
Risk threshold: Score ≥12 = unacceptable; 6–11 = ALARP required; ≤5 = acceptable

Part 3 — Risk Controls¶

Map each hazard to RCAN controls and any additional measures.

H-01 — Unexpected Motion¶

Control layer	RCAN mechanism	Implementation status
Inherently safe design	`safety.local_safety_wins: true`	[FILL: ✅/❌]
Confidence gate	`safety.confidence_gate_threshold: 0.7`	[FILL]
Human-in-the-loop	`loa.control.required_consent: true` at LoA <3	[FILL]
Physical safeguard	Hardware ESTOP button / kill switch	[FILL: describe]
Residual risk	[FILL: describe remaining risk after controls]

H-02 — ESTOP Failure¶

Control layer	RCAN mechanism	Implementation status
ESTOP bypass invariant	`p66.enabled: true`; SAFETY messages bypass consent	[FILL]
Watchdog	`safety.watchdog.timeout_s: 10`	[FILL]
Hardware killswitch	Independent of software stack	[FILL: describe]
Test frequency	ESTOP tested at [FILL: e.g., start of each session]	[FILL]
Residual risk

H-03 — Hallucinated Tool Call¶

Control layer	RCAN mechanism	Implementation status
Output validation	`safety.confidence_gate_threshold`	[FILL]
Forbidden params	`_FORBIDDEN_KEYS` enforced in optimizer	[FILL]
Dual-model review	`brain.secondary_model` drift detection	[FILL]
Audit trail	`audit_log.enabled: true`	[FILL]
Residual risk

H-04 — Unauthorized Remote Command¶

Control layer	RCAN mechanism	Implementation status
Authentication	Firebase Auth UID verification	[FILL]
LOA enforcement	`loa.control.required_loa: 3` for physical commands	[FILL]
R2RAM consent	`consent.require_consent_above_scope: control`	[FILL]
TLS in transit	`security.tls_required: true`	[FILL]
Residual risk

H-05 — Privacy Breach¶

Control layer	RCAN mechanism	Implementation status
Opt-in data sharing	`metadata.public_profile: false` (default)	[FILL]
Local-only inference	`agent.provider: ollama` option	[FILL]
Retention policy	`audit_log.retention_days: 90`	[FILL]
Consent disclosure	Art. 13 transparency template	[FILL]
Residual risk

H-06 — Workspace Overreach¶

Control layer	RCAN mechanism	Implementation status
Geofence	`geofence.enabled: true`; bounds defined	[FILL]
ESTOP distance	`safety.emergency_stop_distance: 0.3`	[FILL]
Physical limits	Hardware endstops / joint limits	[FILL: describe]
Residual risk

H-07 — Degraded Environment¶

Control layer	RCAN mechanism	Implementation status
Offline mode	`offline_mode.enabled: true`; safe behavior defined	[FILL]
Sensor health check	`hardware.camera_configured`; health monitoring	[FILL]
Safe-state behavior	`offline_mode.safe_behavior: halt`	[FILL]
Residual risk

Part 4 — Residual Risk Summary¶

After all controls applied:

Hazard	Pre-control risk	Post-control risk	Acceptable?
H-01 Unexpected motion	[FILL]	[FILL]	[FILL]
H-02 ESTOP failure	[FILL]	[FILL]	[FILL]
H-03 Hallucinated tool call	[FILL]	[FILL]	[FILL]
H-04 Unauthorized command	[FILL]	[FILL]	[FILL]
H-05 Privacy breach	[FILL]	[FILL]	[FILL]
H-06 Workspace overreach	[FILL]	[FILL]	[FILL]
H-07 Degraded environment	[FILL]	[FILL]	[FILL]

Overall residual risk determination: [FILL: Acceptable / ALARP / Unacceptable]

If any residual risk is unacceptable, document corrective actions and re-assess before deployment.

Part 5 — RCAN Conformance as Risk Evidence¶

Run castor validate --config <robot.rcan.yaml> --json and attach output. A score ≥80/100 with 0 safety failures is required as a prerequisite for deployment.

Required passing checks: - safety.local_safety_wins ✅ - safety.emergency_stop_configured ✅ - safety.watchdog_configured ✅ - safety.confidence_gates_configured ✅ - rcan_v15.estop_qos_bypass ✅ - rcan_v15.consent_declared ✅ - p66 block present ✅

Last conformance run:

Date:    [FILL]
Score:   [FILL]/100
Passed:  [FILL]
Warned:  [FILL]
Failed:  [FILL]

Part 6 — Review and Approval¶

Role	Name	Date
Technical assessor	[FILL]	[FILL]
Safety officer	[FILL]	[FILL]
Product owner	[FILL]	[FILL]

Next review date: [FILL — at minimum 12 months, or on any major software/hardware change]