Automate CE Step 5

5. Perform risk assessment, design controls, and testing

High-level guidance 

Objective: Perform a human‑led risk assessment aligned with ISO 12100 and relevant harmonised standards.
AI may support auxiliary tasks only (retrieval, structuring, cross‑checking), while all core safety analysis—hazard identification, causation analysis, risk evaluation, and selection of risk‑reduction measures—must be performed by competent humans.

Human responsibilities (core safety activities)

• Identify hazards

• Analyse causes and sequences of events

• Determine harms

• Evaluate risk (initial and residual)

• Select risk‑reduction measures

• Decide required tests

• Judge acceptability of residual risk

AI responsibilities (auxiliary, non‑safety‑critical)

• Retrieve relevant directives, regulations, standards

• Summarise clauses without interpreting them

• Check completeness of human‑identified hazard categories

• Check consistency of hazard → risk → mitigation → verification chains

• Format documentation

• Highlight missing regulatory references

• Compare versions of documents

High‑level prompting techniques (with examples)

1. Chain‑of‑thought prompting

Used only to check structure and completeness of human‑created content. Enforce structural completeness without generating safety content.

Example prompt:

“Review my human‑created risk table and check whether each hazard has a complete chain:
hazard → hazardous situation → harm → mitigation → verification.
Identify only missing links. Do not add hazards, risks or controls.”

2. Atom‑of‑thought prompting

Used to categorise human‑identified hazards into ISO 12100 categories. Structure and traceability, not hazard creation.

Example prompt:

“Categorise each of these hazards into ISO 12100 hazard groups (mechanical, electrical, thermal, etc.).
Do not create new hazards.”

3. Self‑checking prompting

Used to detect omitted categories, not to generate hazards. Ensure completeness of categories, not content.

Example prompt:

“Compare my hazard list with ISO 12100 hazard categories.
List only which categories are missing.
Do not propose hazards.”

4. Few‑shot prompting

Used to align format and style with ISO 12100 / EN 62368‑1 / EN 71 examples. Stylistic alignment, not safety analysis.

Example prompt:

“Using the structure shown in these ISO 12100‑style examples, re‑format my human‑written hazard table.
Do not change the meaning or add content.”

Detailed guidance can be purchased below 🔽