Skip to content

Use AI Tools Responsibly

You have access to AI tools like ChatGPT, Copilot, or GovSafeAI. This journey helps you use them effectively while keeping yourself and your organisation safe.

The Golden Rules
  • Never trust blindly - AI makes confident-sounding mistakes
  • Never input sensitive data - Unless you're certain it's allowed
  • Always verify important outputs - You're accountable, not the AI
  • Know your policies - Different organisations have different rules

Before You Start

Check What's Allowed

Your organisation likely has rules about AI use. Find out:

  • What AI tools are approved for use?
  • What can and can't be input?
  • What use cases are permitted?
  • Who do you ask if unsure?

If in doubt, ask

Using unapproved tools or inputting sensitive data can have serious consequences.

Understand the Basics

What AI can do well: - Draft content you'll review - Explain concepts - Summarise long documents - Generate ideas and options - Help with coding - Translate languages

What AI does poorly: - Facts and citations (often invents them) - Current events (knowledge may be outdated) - Complex reasoning - Anything requiring real expertise - Understanding your specific context - Being consistent

Using AI Safely

What NOT to Input

Never input these

  • Personal information about citizens or staff
  • Classified or sensitive documents
  • Internal strategies or plans not for sharing
  • Passwords or credentials
  • Proprietary or confidential information
  • Anything you wouldn't want public

Before Inputting, Ask:

  1. Is this data allowed per my organisation's policy?
  2. Would I be comfortable if this appeared publicly?
  3. Does this contain personal information?
  4. Is there a safer way to get the same result?

Using AI for Different Tasks

Good practice:

  • Start with your key points, ask AI to expand
  • Have AI suggest structure, you fill in detail
  • Use as first draft, heavily edit
  • Check all facts independently

Caution:

  • AI may generate plausible but wrong content
  • Tone may not match your needs
  • May include biases or inappropriate content
  • Your name goes on it, not the AI's

Good practice:

  • Use to understand concepts
  • Ask for explanations of complex topics
  • Summarise documents you've provided
  • Generate search terms and questions

Caution:

  • Never trust citations—verify them
  • Knowledge may be outdated
  • May miss nuances or exceptions
  • Can confidently present misinformation

Good practice:

  • Explain code and debug
  • Generate boilerplate
  • Suggest approaches
  • Write tests and documentation

Caution:

  • Generated code may have bugs
  • Security vulnerabilities possible
  • May not follow your standards
  • Always review and test

Good practice:

  • Generate lists of options
  • Challenge your thinking
  • Explore different perspectives
  • Overcome blank page syndrome

Caution:

  • May suggest impractical ideas
  • Can reinforce biases
  • Quality varies widely
  • Still need human judgment

Getting Good Results

Prompting Tips

Technique Example
Be specific "Write a 200-word summary for executives" vs "Summarise this"
Provide context "I'm a policy officer in health. Help me..."
Specify format "Respond in bullet points" or "Give me a table"
Ask for alternatives "Give me three different approaches"
Iterate "Make it shorter" or "Add more detail on X"

Quality Verification

Before using any AI output:

Check Why
Facts AI invents plausible-sounding facts
Sources Citations are often fabricated
Logic Reasoning may be flawed
Tone May not match your needs
Completeness May miss important considerations
Bias May reflect or amplify biases

When to Double-Check

Always verify when:

  • The output will be seen by others
  • Decisions will be based on it
  • It contains statistics or claims
  • It cites sources or references
  • Accuracy really matters
  • It doesn't feel quite right

When Things Go Wrong

Recognising Problems

Signs of AI errors:

  • Very specific numbers or dates (often invented)
  • Confident statements on contentious topics
  • Citations that look almost right
  • Content that seems too good to be true
  • Inconsistent information within response

What to Do

If AI produces something wrong:

  1. Don't use the incorrect output
  2. Fact-check before proceeding
  3. Adjust your prompt and try again
  4. Consider if AI is right for this task

If AI produces something harmful:

  1. Don't share or use the output
  2. Report per your organisation's process
  3. Document what happened
  4. Consider if policy changes needed

If you accidentally input sensitive data:

  1. Stop immediately
  2. Report to your IT/security team
  3. Don't try to "undo" it—you can't
  4. Document what was input

Building Good Habits

Daily Practices

  • Pause before inputting—is this data appropriate?
  • Verify before sharing—have you checked the output?
  • Attribute appropriately—be transparent about AI use
  • Learn continuously—AI tools change rapidly

Questions to Ask Yourself

Before using AI output:

  1. Have I verified the facts?
  2. Would I defend this work as my own?
  3. Does this meet quality standards?
  4. Am I being transparent about AI use?

When NOT to Use AI

Situation Why
Legal advice Need qualified professionals
Medical decisions Potentially dangerous errors
Personnel matters Sensitivity, bias risks
Security decisions Too important to risk
Where prohibited Follow your policies

Learning More

Build your skills:

  • Practice prompting techniques
  • Learn from what works and what doesn't
  • Share tips with colleagues
  • Stay updated on new capabilities
  • Attend training when available

Stay informed:

  • AI capabilities change rapidly
  • Policies update over time
  • New risks emerge
  • Best practices evolve

Quick Reference Card

✅ Do

  • Use approved tools only
  • Verify important outputs
  • Think before inputting data
  • Ask if unsure
  • Be transparent about AI use
  • Report problems

❌ Don't

  • Input sensitive data
  • Trust outputs blindly
  • Skip verification steps
  • Use for prohibited purposes
  • Pretend AI work is entirely yours
  • Ignore your policies

🆘 Help

  • Policy questions → Your manager / IT
  • Technical issues → IT support
  • Security concerns → Security team
  • Something went wrong → Report immediately