Law OF AI: Bias, Evidence, Liability, IP, Data Protection
Stop treating AI only as a tool. Now treat it as a subject of law — and watch how that makes you a safer user of it.
The hook
“A risk-assessment algorithm helped decide a man's sentence — and he was never allowed to see how it scored him. Soon the "evidence" against your client may be a video that never happened. The law of AI is no longer speculative; it is your next brief.”
What you'll be able to do
- Treat AI as a subject of law, not just a tool, and acquire the law-and-technology literacy expected of an NLU graduate.
- Analyze the core law-of-AI problem clusters at an informed level: algorithmic bias and due process, AI-fabricated evidence and authentication, IP and copyright in training data, and data protection under the DPDP Act 2023.
- Map the regulatory landscape — the EU AI Act, India's emerging AI governance, and the SC White Paper's principles of transparency, fairness, accountability, and human oversight.
- Connect substantive law-of-AI knowledge back to responsible day-to-day use: knowing where the law constrains, forbids, or scrutinizes AI tells you where to use it carefully.
In short
This module flips the course's second track into focus: students stop studying AI only as a tool they prompt and start studying it as a subject of law they will litigate, regulate, and advise on. Across bias and due process, evidence and deepfakes, IP in training data, and data protection under the DPDP Act, students build the law-and-technology literacy NLU graduates are expected to have — and see that understanding the law OF AI is what makes you a more responsible user OF AI.
The AI bridge
Understanding the law OF AI makes you a better, safer user OF AI — knowing where bias, evidence rules, IP, and data-protection law constrain or forbid reliance on a model tells you exactly where to verify hardest and where not to rely at all. It is also, in itself, a growth practice area.
In this module
- 01
Algorithmic bias and due process: the COMPAS risk-assessment tool and ProPublica's 2016 investigation into racial disparities, and State v. Loomis (Wisconsin) — where a defendant challenged the use of an opaque risk score in sentencing on due-process grounds. The North-Star tie: when you cannot see how a model reached a score, you cannot verify or interrogate it — the same blind spot that makes a confident LLM dangerous in your own hands.
- 02
Predictive policing as the same bias problem moved upstream: models trained on historical enforcement data can launder past discrimination into future decisions. The lesson for the responsible user: a model's output inherits the biases of its data, so where stakes are high you must scrutinize provenance, not just fluency.
- 03
Evidence law and deepfakes: authentication of AI-fabricated evidence, the admissibility problem, and the 'liar's dividend' — the corrosive effect where the mere existence of deepfakes lets genuine evidence be dismissed as fake. The tie: a lawyer who understands how synthetic media is generated knows where and how hard to demand authentication.
- 04
IP and copyright in training data and AI output: NYT v. OpenAI and the Indian ANI v. OpenAI suit raise whether training on copyrighted material infringes, plus open authorship and ownership questions over machine-generated output. The responsible-use payoff: knowing the IP exposure of AI-generated work product is part of competent client advice.
- 05
Data protection under India's DPDP Act 2023: limits on automated decision-making and the contested 'right to explanation'. This grounds the confidentiality discipline the course teaches — why client and personal data cannot be casually fed into public models.
- 06
The regulatory landscape: the EU AI Act as the comparative frame, India's emerging AI governance, and the SC White Paper's four principles — transparency, fairness, accountability, and human oversight. These principles are the checklist a responsible AI-using lawyer internalizes.
- 07
The synthesis beat: understanding the law OF AI makes you a better, safer user OF AI — you use AI well precisely because you know where the law constrains it — and the law of AI is itself a fast-growing practice area for these students' careers.
The interactive demos
Every idea is a Mirror Move
Run it on the room, show it inside the machine, prove it live on a real AI, then name the skill.
Real or AI? — the deepfake authentication problem
On us
Run the 'real or AI?' deepfake set on the room: show a sequence of images/clips and have students commit a confidence vote on which are genuine and which are AI-generated before any reveal.
In the machine
Generative models produce synthetic media that is fluent and confident-looking yet has no connection to any real event — the same way an LLM produces confident text untethered from truth.
Live AI
Reveal the answers and surface the confidence gap: the room was often sure about items it got wrong, demonstrating how persuasive synthetic media is and how the 'liar's dividend' lets genuine evidence be doubted.
The skill
Treat this as an evidence-authentication problem: never accept media (or an AI answer) on the strength of how convincing it looks — demand provenance and authentication proportional to the stakes.
The lab
Issue-spot the AI
Students are given a short AI-deployment fact pattern and must issue-spot the legal risks it raises across the module's four clusters — bias, privacy/data protection, IP, and evidence — the same way they would issue-spot any moot problem, but now with AI as the subject of law.
Deliverable
A written issue-spotting analysis identifying the bias, privacy, IP, and evidence risks in the fact pattern, with each risk tied to the relevant law-of-AI source or principle.
Key sources & cases
State v. Loomis, 881 N.W.2d 749 (Wis. 2016)
Wisconsin risk-assessment / due-process case; defendant challenged use of an opaque COMPAS risk score in sentencing. Anchor for the algorithmic-bias-and-due-process cluster.
ProPublica COMPAS investigation (2016)
Investigation into racial disparities in the COMPAS recidivism risk-assessment tool; the empirical backbone of the bias-and-due-process discussion.
NYT v. OpenAI (pending)
US suit raising whether training LLMs on copyrighted material infringes; primary IP / copyright-in-training-data example.
ANI v. OpenAI (Delhi HC, pending)
The Indian copyright suit over training data; the India-first counterpart to NYT v. OpenAI for the IP cluster.
DPDP Act 2023 (India)
India's Digital Personal Data Protection Act; frames automated decision-making and the contested 'right to explanation'; grounds the confidentiality discipline.
EU AI Act
Comparative regulatory frame for the law-of-AI landscape alongside India's emerging governance.
Supreme Court of India White Paper on AI and the Judiciary (Nov 2025)
Source of the four governing principles — transparency, fairness, accountability, human oversight — used as the responsible-use checklist.
Cathy O'Neil, Weapons of Math Destruction
Core reading on opaque, biased algorithmic decision systems; underpins the bias cluster.
Hannah Fry, Hello World
Accessible reading on algorithms in society; supports the bias and decision-system material.
Frank Pasquale, The Black Box Society and New Laws of Robotics
Law-and-technology readings on algorithmic opacity and the regulation of automated systems; frame the module's legal analysis.
Readings
- Cathy O'Neil, Weapons of Math Destruction (2016)
- Hannah Fry, Hello World (2018)
- Frank Pasquale, The Black Box Society (2015) and New Laws of Robotics (2020)
- State v. Loomis, 881 N.W.2d 749 (Wis. 2016)
- ProPublica COMPAS investigation (2016)
- NYT v. OpenAI (pending) and ANI v. OpenAI (Delhi HC, pending)
- India's DPDP Act 2023
- The EU AI Act
- Supreme Court of India White Paper on AI and the Judiciary (Nov 2025)
Next module
Module 08 / 08
Capstone & the Future Lawyer: Becoming the Human-in-the-Loop
synthesizeBring a credit-bearing AI course to your students
A 16-hour course that treats using AI well as a professional duty — one a council can approve, and a graduate can defend in court.