Framework
The STOP Framework
When Should You NOT Use AI? A four-checkpoint assessment — Security, Trust, Ownership, Purpose.
The STOP Framework is a systematic assessment methodology for evaluating AI use cases before implementation. Each letter represents a critical question that must be answered honestly before proceeding.
By asking these four questions in sequence, teams can identify high-risk situations before they cause harm—turning reactive compliance into proactive governance.
S
Security Check: Does it Handle Sensitive Data?
What to consider
- Personal information (names, addresses, identifiers)
- Confidential business data (strategy, financials, IP)
- Regulated data (health, financial, legal)
- Third-party confidential information
Red flags
- Client data being processed by external AI services
- Sensitive documents being uploaded to cloud AI tools
- Personal information in prompts or training data
If yes: Implement data governance protocols, consider on-premise solutions, establish clear data retention policies.
T
Trust Risk: What Happens If It Hallucinates?
What to consider
- Impact on decisions being made with AI output
- Consequences of inaccurate information
- Visibility of AI use to stakeholders
- Professional/legal implications of errors
Red flags
- AI output going directly to clients without review
- Using AI for legal, medical, or financial advice
- AI making autonomous decisions in high-stakes contexts
If high risk: Implement human review processes, establish verification protocols, create clear escalation paths.
O
Ownership: Who Is Accountable for the Output?
What to consider
- Clear accountability for AI-assisted work
- IP ownership of AI-generated content
- Liability for errors or harm
- Attribution requirements
Red flags
- No clear human owner for AI outputs
- Ambiguous responsibility chains
- “The AI did it” as a defence
If unclear: Establish clear ownership protocols, document human oversight requirements, create accountability frameworks.
P
Purpose: Is AI the Right Tool for This Task?
What to consider
- Does this task actually benefit from AI?
- Are there simpler, more reliable alternatives?
- Does the complexity justify the risk?
- Is this enhancing or replacing human judgment inappropriately?
Red flags
- Using AI for tasks better suited to traditional tools
- AI as a solution looking for a problem
- Automating tasks that require human judgment
If questionable: Consider alternatives, pilot with low-stakes use cases, measure actual value added.