AI Hallucinations and Legal Ethics: Managing the Risk

The Hallucination Problem

AI "hallucination" is a polite term for AI making things up. It's not a bug—it's a feature of how large language models work. They predict plausible text, and sometimes plausible isn't true.

In legal contexts, this creates real professional risk. Cases that don't exist, statutes that were never enacted, precedents with incorrect holdings—AI can generate all of these with complete confidence.

The US has seen multiple high-profile cases where lawyers cited AI-generated case law that turned out to be fabricated. UK regulators and courts are watching closely.

Why AI Hallucinations Happen

Understanding the mechanism helps manage the risk:

How Language Models Work

Large language models don't "know" facts like a database. They predict what text should come next based on patterns learned from training data. When asked a legal question, the model generates text that looks like a legal answer—whether or not the content is accurate.

Confidence Without Knowledge

AI doesn't have uncertainty in the human sense. It generates text with the same confidence whether the content is:

• Clearly established law
• Disputed or uncertain
• Completely made up

There's no warning label on fabricated content.

The Specificity Trap

Paradoxically, AI can be more wrong when it's more specific:

• Asked for "a case about contract formation," it might find a real one
• Asked for "a 2019 Court of Appeal case about email formation of contracts," it might fabricate one that fits the specification

Real-World Examples

US Cases

Multiple US lawyers have faced sanctions for citing AI-fabricated cases:

• New York attorney fined $5,000 for fake citations in federal court
• Texas attorney sanctioned after AI-generated brief contained non-existent cases
• Several other cases in various jurisdictions

UK Implications

While no UK cases of AI citation sanctions have been widely reported yet, the risk is identical:

• SRA would likely treat it as competence failure
• Courts would not be sympathetic
• Professional reputation damage would be severe

Appropriate Use Cases

AI can be helpful for legal work when used appropriately:

Lower Risk Uses

Higher Risk Uses

Verification Protocols

For any AI-generated legal content, establish verification procedures:

Case Citations

Every citation must be verified:

1. Check the case exists in a legal database (Westlaw, LexisNexis, BAILII)
2. Verify the citation format is correct
3. Confirm the holding matches what AI claimed
4. Check the current status (not overruled, distinguished)

If you can't verify it, don't cite it.

Statutory References

For legislation:

1. Confirm the statute exists
2. Check the section/provision is real
3. Verify the wording matches (AI paraphrases can subtly change meaning)
4. Check for amendments since AI training data

Procedural Statements

For practice and procedure:

1. Verify against current rules
2. Check practice directions
3. Confirm local variations if applicable
4. Be alert to outdated procedures

General Statements of Law

For legal propositions:

1. Consider whether it sounds correct based on your knowledge
2. If uncertain, research independently
3. Don't rely on AI for areas outside your competence
4. Document your verification

Training Staff on AI Risks

Everyone using AI needs to understand:

The Nature of AI

• AI generates plausible text, not verified truth
• Confidence doesn't indicate accuracy
• Specific details can be entirely fabricated
• Legal content requires verification

Professional Obligations

• You're responsible for work product, regardless of how generated
• SRA competence requirements apply
• Courts expect accurate citations
• "AI told me" is not a defence

Practical Protocols

• Always verify citations before use
• Use approved tools and methods
• Escalate if uncertain
• Document verification performed

Red Flags to Watch For

• Cases with suspiciously perfect facts
• Obscure courts or reporters
• Citations that can't be found
• Holdings that seem too good to be true

Managing the Workflow

Separation of Drafting and Verification

Don't verify as you draft—it's easy to skip or rush:

1. Use AI for initial draft
2. Complete the draft
3. Perform systematic verification
4. Document verification completed

Verification Checklist

For legal content, verify:

• All case citations checked in legal database
• All statutory references confirmed
• General legal statements validated
• Procedural points checked against current rules
• Content within my competence to verify

Time Budget for Verification

Verification takes time. AI saves drafting time but adds verification time:

• Factor verification into matter planning
• Don't assume AI makes everything faster
• Budget time for independent research if needed

The Competence Question

A fundamental question for AI use:

Can you verify the AI output within your competence?

If AI generates content about employment law discrimination tests and you don't practice employment law:

• You can't verify whether the law stated is correct
• You can't spot subtle errors in application
• Using it would be practicing outside your competence

AI doesn't extend your competence—it amplifies what you can do within it.

When AI Isn't Appropriate

Be cautious using AI for:

Unfamiliar Areas

If you wouldn't feel confident drafting it yourself, AI assistance doesn't change that.

High-Stakes Matters

When errors have severe consequences, the verification burden may outweigh AI benefits.

Time-Pressured Situations

When there isn't time for proper verification, AI shortcuts become dangerous.

Complex, Novel Issues

AI trained on historical data may not reflect recent developments or novel situations accurately.

Professional Responsibility Bottom Line

The SRA has been clear: you remain responsible. Using AI tools doesn't delegate your professional judgment or liability.

Firms that use AI well will:

• Understand its limitations
• Verify outputs systematically
• Use it within their competence
• Document their processes
• Train staff appropriately

Firms that use AI poorly will eventually face:

• Regulatory action
• Professional negligence claims
• Court sanctions
• Reputational damage

The choice is how you use it, not whether to use it.

---

Need guidance on managing AI risk in your practice? We help law firms implement AI tools with appropriate safeguards and verification protocols.

Book a consultation to discuss your specific needs.