Workshop

12 AI Failure Modes: How LLMs Think, Fail & Mislead in Legal Work

5.0 (1)

Tatiana Botskina

Oxford DPhil CS | Legal AI Researcher

Equip yourself with deep AI knowledge to use AI responsibly in legal work

Equip yourself with deep AI knowledge to use AI responsibly in legal work

This two-hour intensive introduces 12 types of AI errors in legal work that every lawyer needs to recognise and provides the technical foundation to use AI safely, evaluate its outputs critically, and know when not to trust it.

Built on five years of doctoral research at the University of Oxford into generative AI, explainability, LLM failure modes and transformer architectures, the session goes far beyond hallucinations to examine the less obvious failure modes emerging in today's most advanced AI systems.

You will learn why AI can miss information that is directly in front of it, overlook clauses in long documents, follow some instructions while silently ignoring others, and produce correct answers for the wrong reasons.

Through 12 distinct error types and practical legal examples, you will learn what causes these failures, why they can be difficult to detect, and where they create particular risks in legal work.

Most importantly, you will learn how to detect, evaluate and mitigate AI failures and exercise informed professional judgement when using AI in legal practice.

What you’ll learn

Essential training for any lawyer who relies on AI: understand how it fails, verify its outputs, and use it safely in legal practice.

  • moving beyond hallucinations to recognise reasoning, context, instruction-following, sycophancy and agentic failure modes

  • determining what must be checked, when human review is necessary and where AI should not be relied upon

  • using a technical framework rather than intuition, vendor claims or generic accuracy scores

Workshop agenda

  • Technical Introduction into LLMs

    A lawyer-accessible introduction to transformer architecture: tokens, attention, representations, prediction and why understanding the underlying mechanism matters for legal work.

  • Beyond Hallucinations: Modern AI Failure Mode

    Hallucinations and fabricated authorities; context omission; long-document failures; instruction-following failures; reasoning errors; sycophancy and deception; inconsistency and overconfidence;

  • When AI Becomes an Agent

    How failures change when LLMs use tools and execute multi-step tasks

  • How to Evaluate AI Output

    A practical framework for distinguishing plausible output from reliable output. Why benchmarks, citations, confidence and self-checking can be misleading and what lawyers should evaluate instead.

Learn directly from Tatiana

Tatiana Botskina

Tatiana Botskina

Contact

Oxford DPhil CS candidate (Legal AI) | AI Executive | 5x Founder | Forbes 30U30

Previously at
University of Oxford
Norton Rose Fulbright
See all products from Intelliigend AI

Who this workshop is for

  • Senior lawyers who use AI tools daily but feel uneasy, you're not sure when the output is reliable.

  • Legal leaders being asked to approve AI budgets and set firm-wide AI policies without a real understanding of how the technology works.

  • In-house counsel and legal ops professionals under pressure to "adopt AI" while protecting corporate data and managing regulatory exposure.

What's included

Certificate of completion

Share your new skills with your employer or on LinkedIn.

Maven Guarantee

Your purchase is backed by the Maven Guarantee.

Frequently asked questions

Maven for Teams

Reimbursement

Get your company to pay

Everything L&D needs: email template, receipts, and certificate of completion.

Get reimbursed

Private cohort

Run a cohort for your org

A dedicated cohort with a custom schedule and curriculum, tailored to your team.

Book a private cohort