AI Safety & Ethics

Research-backed practices for responsible AI use. Understanding risks and building safe habits for human-AI collaboration.

AI for Everybody Built With UD/UDL Security A+ 100% Performance 100% AI Assisted Building Claude Code Community GitHub

Understanding AI Risks

Research-identified risks every AI user should understand

High Frequency

Hallucinations

AI can generate convincing but entirely fictional information, including fake citations, invented statistics, and fabricated facts. Hallucination rates vary by task complexity.

Warning Signs

  • Specific numbers without verifiable sources
  • Confident claims about obscure topics
  • Citations that can't be found online
  • Details that seem "too perfect"
Practice Spotting Hallucinations
Systemic

Bias & Fairness

AI systems can reflect and amplify biases present in training data. These biases may be cultural, gender, racial, or socioeconomic, and often appear in subtle ways.

Mitigation Strategies

  • Test outputs across diverse scenarios
  • Question assumptions in AI responses
  • Use neutral prompt language
  • Seek multiple perspectives on outputs
Try Hallucination Spotter
Privacy

Data Exposure

Information shared with AI systems may be stored, analyzed, or used for training. Privacy risks increase when users share sensitive personal or organizational data.

Never Share

  • Passwords or authentication credentials
  • Personal identifiers (SSN, medical IDs)
  • Confidential business information
  • Private communications of others
Behavioral

Over-Reliance

Automation bias can lead users to accept AI outputs without critical evaluation. This erodes human skills and judgment over time, creating dependency.

Maintain Balance

  • Use AI to augment, not replace, thinking
  • Practice skills independently sometimes
  • Always review before accepting outputs
  • Question confident-sounding responses

The VERIFY Technique

A research-backed approach to evaluating AI outputs

V - Validate Sources

Check if cited sources actually exist and say what the AI claims. Hallucinated citations are common.

E - Examine Logic

Does the reasoning make sense? Look for logical gaps, unsupported leaps, or circular arguments.

R - Research Further

Cross-reference important claims with reliable independent sources. One source isn't enough.

I - Inspect for Recency

AI training data has cutoff dates. Check if information might be outdated for your needs.

F - Flag Bias

Consider if the response shows cultural, political, or other biases that could skew accuracy.

Y - Your Judgment

Apply your own expertise and common sense. You are the final quality control.

Real-World Scenarios

Learn from common AI safety situations

Scenario

The Confident Citation

Situation: You ask AI to provide research supporting a business proposal. It returns three academic citations with authors, journals, and publication years that sound authoritative.

The Risk: Two of the three citations are completely fabricated. The journal names exist, but those specific papers don't. The AI hallucinated convincing-sounding academic references.

Safe Practice: Before using any citation, search for it directly in Google Scholar or the journal's website. If you can't find the exact paper, it likely doesn't exist.

Lesson: AI can generate academically-styled citations that don't exist. Always verify before citing.
Scenario

The Biased Recommendation

Situation: You ask AI to help write job requirements for a technical role. The output consistently uses masculine pronouns and emphasizes "aggressive" and "dominant" traits.

The Risk: This language can discourage diverse candidates from applying and perpetuate workplace homogeneity. The bias reflects patterns in training data.

Safe Practice: Review AI-generated professional content for gendered language, cultural assumptions, and unnecessarily exclusionary requirements. Use the Bias Radar tool to catch subtle issues.

Lesson: AI reflects societal biases. Human review ensures inclusive, fair outputs.
Scenario

The Privacy Leak

Situation: An employee pastes a customer complaint email into AI to draft a response. The email contains the customer's full name, address, phone number, and account details.

The Risk: Personal information has now been shared with a third-party AI system. Depending on the AI provider's terms, this data might be stored, analyzed, or used for training.

Safe Practice: Remove or redact personal identifiers before sharing content with AI. Use placeholders like [CUSTOMER_NAME] and [ACCOUNT_NUMBER].

Lesson: Treat AI like a public conversation. Don't share what you wouldn't post publicly.

High-Stakes Domains

Areas requiring extra caution and human expertise

Critical Principle

In high-stakes domains, AI should inform human decision-making, never replace it. The consequences of AI errors in these areas can be severe and irreversible.

Medical & Health

AI is not a doctor. Never make health decisions based solely on AI. Always consult qualified healthcare professionals for diagnosis and treatment.

Legal Matters

AI cannot provide legal advice. Laws vary by jurisdiction and situation. Consult licensed attorneys for legal guidance.

Financial Decisions

AI shouldn't make investment or major financial decisions. Consult qualified financial advisors for personalized guidance.

Safety-Critical Systems

Decisions affecting physical safety, infrastructure, or life-and-death situations require human expertise and accountability.

Daily Safety Practices

Habits that reduce AI-related risks

Practice Why It Matters How to Apply
Verify before trusting Hallucinations are common and convincing Cross-reference facts, check citations exist
Review before sharing AI outputs may contain errors or bias Read everything before sending or publishing
Protect privacy Data shared may be stored or used Redact personal info, use placeholders
Maintain your skills Over-reliance erodes judgment Do some tasks without AI regularly
Disclose AI use Transparency builds trust Mention AI assistance when appropriate
Stay informed AI capabilities change rapidly Follow updates from AI providers

Practice Safe AI Use

Build your verification skills with these tools

Hallucination Spotter

Practice identifying AI-generated misinformation

Persona Architect

Build expert AI personas for consistent responses

Preflight Checklist

Verify prompts before sending

Sources