L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Module 2 · Lesson 1

The World's First AI Law

How Europe decided to regulate AI — and what that means for everyone

After three years of drafting and negotiating, the European Parliament voted 523–46 to pass the EU AI Act. Reporters called it historic. Critics called it a blunt instrument aimed at a fast-moving target.

The Act's core premise: AI systems that make consequential decisions about people's lives deserve scrutiny proportional to the risk they create.

The Risk-Based Architecture

The EU AI Act uses a four-tier risk framework. Unacceptable risk systems are banned outright — government social scoring, real-time biometric surveillance in public spaces. High-risk systems in critical domains face full compliance requirements. Limited-risk systems like chatbots need transparency disclosures. Minimal-risk systems have no specific obligations.

The Act applies to any AI deployed in the EU — regardless of where its developer is based. Like GDPR, it operates extraterritorially.

Why Risk-Tiering?

Matching compliance burden to harm potential — the same principle as product safety law. Medical devices face stricter testing than kitchen appliances not because regulators distrust kitchens, but because the consequences of failure differ dramatically.

The High-Risk Compliance Stack

High-risk systems — AI in hiring, credit, law enforcement, healthcare, education, critical infrastructure — must meet demanding requirements: an ongoing risk management system, data governance standards, full technical documentation, transparency obligations, human oversight design (systems must not circumvent human review), and accuracy and robustness standards. Both providers (who build) and deployers (who use) have distinct obligations. Penalties reach €35 million or 7% of global annual turnover.

General-Purpose AI Models

A late addition covers foundation models like GPT-4 and Claude. All GPAI providers must publish training data summaries and usage policies. Models trained above 10^25 FLOPs — the frontier tier — must additionally conduct adversarial testing, report serious incidents, and track energy use. This creates layered liability chains: model provider, application developer, and deployer each have their own obligations — and each depends on the one above for documentation.

The Brussels Effect

Historically, EU regulations shape global standards as companies build one product meeting the strictest requirements rather than jurisdiction-specific variants. The EU AI Act is expected to influence AI governance worldwide.

Lesson 1 Quiz

EU AI Act foundations
The EU AI Act applies to:
✓ Correct — Correct. The Act is extraterritorial — it applies based on where people are affected, not where developers are headquartered.
The Act applies to any AI deployed in the EU — including systems from US, Chinese, or any non-EU companies.
Government social scoring is classified under the EU AI Act as:
✓ Correct — Correct. Social scoring is in the prohibited tier — banned entirely as incompatible with fundamental rights.
Government social scoring is in the unacceptable risk tier and is prohibited outright, not merely regulated.
The Act's risk-tiering approach was designed to:
✓ Correct — Correct. Risk-tiering is about proportionality — matching compliance burden to harm potential.
Risk-tiering concentrates compliance burden where consequences of failure are most serious. Low-stakes AI faces light or no obligations.
Under the EU AI Act, "deployers" are:
✓ Correct — Correct. Deployers use AI in their operations — they have compliance obligations separate from the provider.
Deployers are organizations using AI systems. Providers are those who build and sell them.

Lab 1 — Risk Tier Classification

Apply the four-tier framework to real AI systems

Your Task

Choose an AI system (or pick one: predictive policing, university admissions, workplace productivity tracker, medical symptom checker, social media feed, credit scoring model).

Classify it under the four risk tiers. Justify with specific harms, affected parties, and decision types.

Name your system and state your tier classification — then defend it.
AI Lab AssistantEU AI Act Risk Classifier
Ready. Name your AI system and give me your initial tier classification. I will test your reasoning.
Module 2 · Lesson 2

High-Risk Compliance

What the Act actually requires of AI systems that make consequential decisions

The hiring manager had used the AI screening tool for six months before anyone asked how it worked. She did not know. The vendor had called it "bias-free." She had trusted that.

Now an employment lawyer was asking whether she had the technical documentation the EU AI Act required her to maintain. She had a contract and a user guide. Under the Act, that was not enough.

The Six Core High-Risk Requirements

(1) Risk Management System — documented and ongoing throughout the system's lifecycle, not a one-time assessment. (2) Data Governance — training data must be relevant, representative, and examined for discriminatory patterns. (3) Technical Documentation — sufficient for regulators to assess compliance: design, architecture, training methodology, accuracy metrics. (4) Transparency — deployers receive documentation on capabilities, limitations, and oversight procedures. (5) Human Oversight — systems must be designed so humans can understand outputs, intervene, override, or shut down; must not circumvent oversight. (6) Accuracy, Robustness, Cybersecurity — appropriate accuracy for intended purpose, resilience to attacks.

Conformity Assessment

Before market entry, high-risk systems undergo conformity assessment — self-assessment for most categories, mandatory third-party review for biometric and certain sensitive systems. Approved systems enter an EU registry and receive CE marking. Post-deployment monitoring and incident reporting are also required.

Provider vs. Deployer

Providers (builders) must ensure design meets requirements and maintain documentation. Deployers (users) must use the system within its intended purpose, implement required human oversight, and monitor performance. Both can face enforcement — a company cannot shift all liability by claiming it only purchased the tool.

Penalties

Fines reach €35 million or 7% of global annual turnover — whichever is higher. This puts EU AI Act penalties in the same severity range as major GDPR violations.

Lesson 2 Quiz

High-risk compliance requirements
Human oversight under the EU AI Act means:
✓ Correct — Correct. Human oversight is a design requirement — capability for meaningful supervision, not per-output approval.
Human oversight means the system must be designed to allow meaningful human intervention — it is a design requirement, not approval of every output.
For most high-risk AI systems, conformity assessment is conducted by:
✓ Correct — Correct. Most high-risk systems use provider self-assessment. Third-party review is only required for specific sensitive categories.
Most high-risk systems use provider self-assessment. Third-party conformity assessment is only mandatory for categories like biometric identification.
The maximum EU AI Act penalty for serious violations is:
✓ Correct — Correct. Maximum penalties are comparable in severity to major GDPR violations.
Maximum penalties reach €35 million or 7% of global annual turnover — serious enough to shape large companies strategic decisions.
A deployer violates the EU AI Act by:
✓ Correct — Correct. Deployers must use systems within their documented intended purpose. Using AI outside that scope is a compliance violation.
Deployers violate the Act primarily by using systems outside intended purpose, failing to implement required oversight, or modifying systems in ways creating new risks.

Lab 2 — Compliance Gap Analysis

Identify what a real organization needs to do to comply

Your Task

Choose a specific high-risk scenario: hospital AI triage, bank loan decisions, or court risk-scoring tool.

Go through each of the six high-risk requirements. Assess whether a typical organization currently meets each one — and identify the gap.

State your scenario. Then go requirement by requirement. Be specific — I will challenge any vague claim like "they probably have documentation."
AI Lab AssistantEU AI Act Compliance Analyst
Choose your scenario and start your compliance gap analysis. I will challenge your assumptions about what organizations actually have in place.
Module 2 · Lesson 3

General-Purpose AI Models

The late addition that changed the Act's scope

The original EU AI Act was designed around specific applications — a hiring tool, a medical system, a credit model. Each with a defined purpose and clear provider.

Then came models that could do almost anything. A single model could power hundreds of applications with different risk profiles. Regulators had to go back to the drawing board.

What Are General-Purpose AI Models?

The Act defines GPAI models as AI trained on broad data, capable of performing a wide range of tasks, designed for integration into downstream applications. The paradigm cases: GPT-4, Claude, Gemini, LLaMA. These models have no fixed purpose — their risk depends entirely on how they are deployed. The original Act framework could not handle them and had to be extended in final negotiations.

Baseline GPAI Requirements

All GPAI providers must: Maintain technical documentation (training methodology, data sources, compute, capabilities, safety testing). Comply with EU copyright law by publishing a sufficiently detailed training data summary. Publish a usage policy. Cooperate with downstream providers by sharing documentation they need to meet their own compliance obligations.

Systemic Risk Models

GPAI models trained above 10^25 floating point operations face additional requirements: adversarial testing (red-teaming before deployment and ongoing), serious incident reporting to the European AI Office, cybersecurity measures, and energy consumption reporting. The compute threshold targets frontier models — but critics note training efficiency improvements may make it an unreliable capability proxy over time.

Layered Liability Chains

Foundation model provider, application developer, and deployer each have separate obligations — and each depends on the party above for sufficient documentation. GPAI providers effectively become compliance enablers for their entire downstream ecosystem.

Lesson 3 Quiz

General-purpose AI models
What defines a GPAI model under the EU AI Act?
✓ Correct — Correct. Broad capability and flexible integration define GPAI models — not size, connectivity, or company characteristics.
GPAI models are defined by broad capability and flexible integration into downstream applications, not by parameter count, connectivity, or company size.
The systemic risk tier for GPAI models is triggered by:
✓ Correct — Correct. The FLOPs threshold targets the most computationally intensive frontier models.
The trigger is training compute — 10^25 FLOPs — not users, revenue, or deployment geography.
All GPAI model providers (not just systemic risk tier) must:
✓ Correct — Correct. Training data summaries and usage policies are baseline requirements for all GPAI providers.
All GPAI providers need training data summaries and usage policies. Adversarial testing, incident reporting, and energy tracking are systemic risk tier additions only.
GPAI liability chains exist because:
✓ Correct — Correct. Separate obligations plus documentation dependence creates genuine chains of accountability throughout the AI stack.
Liability chains arise because each layer has independent obligations AND needs upstream documentation to fulfill them — not because liability is shared or transferred.

Lab 3 — GPAI Liability Chain Mapping

Trace compliance obligations through a real AI deployment stack

Your Task

Pick a real AI product built on a foundation model: a legal research tool on GPT-4, a customer service bot on Claude, a coding assistant on LLaMA.

Map all three layers: foundation model provider, application developer, deployer. Assign each layer's specific GPAI obligations.

Name your product and describe all three layers. Then assign obligations to each. I will probe whether your assignments are correct and complete.
AI Lab AssistantGPAI Liability Chain Analyst
Name your AI product and describe the three-layer stack. I will help identify each party's obligations — and challenge any misassignments.
Module 2 · Lesson 4

Enforcement and Critique

Who enforces the Act, what is working, and what critics on both sides get right

The EU AI Act entered into force. The European AI Office was stood up. National market surveillance authorities were designated.

What followed was implementation progress, industry lobbying, civil liberties advocacy, and genuine uncertainty about how the rules would work in practice.

The Enforcement Architecture

Two enforcement levels: National Market Surveillance Authorities handle application-level enforcement in their member states — AI hiring tools, medical systems, credit models. They can investigate, order corrective action, and impose fines. The European AI Office — a new body within the European Commission — supervises GPAI and systemic risk models, coordinates national authorities, maintains the GPAI registry, and can fine providers directly.

Phased Implementation

Requirements phase in over 36 months: prohibited practices at 6 months, GPAI requirements at 12 months, most high-risk requirements at 24 months, some sector-specific high-risk at 36 months. The delays allow standards bodies to develop technical standards and companies to assess their systems.

Civil Liberties Critiques

The law enforcement biometric surveillance exceptions are too broad. Self-assessment for most high-risk systems lets companies grade their own homework. Individual rights to contest automated decisions are inadequate. Prohibited practice definitions have loopholes wide enough for surveillance systems to pass through.

Industry Critiques

Compliance costs disadvantage EU startups against US and Chinese competitors. Legal uncertainty discourages investment. Risk classifications may be miscalibrated. Extraterritorial reach creates friction for global product development.

Scope Gaps

Military and national security AI is explicitly excluded. Personal non-professional activity AI is excluded. Research has broad carve-outs. AI that harms through diffuse societal effects rather than discrete decisions — recommendation algorithms shaping information environments — mostly falls outside the high-risk classification.

Lesson 4 Quiz

Enforcement, implementation, critiques
National Market Surveillance Authorities are primarily responsible for:
✓ Correct — Correct. National MSAs handle application-level enforcement. GPAI model supervision belongs to the European AI Office.
National MSAs enforce application-level compliance in their jurisdictions. GPAI oversight is handled at the EU level by the European AI Office.
Which AI is explicitly outside the EU AI Act's scope?
✓ Correct — Correct. Military and national security AI has an explicit exclusion — a significant gap that critics frequently note.
Military and national security AI is explicitly excluded. Non-EU company systems, critical infrastructure AI, and medical AI are all covered by the Act.
The Brussels Effect means:
✓ Correct — Correct. The Brussels Effect is about regulatory diffusion — EU rules spreading globally through market behavior, not legal force.
The Brussels Effect describes how EU regulations spread globally because companies find it more efficient to meet the strictest standard everywhere.
A civil liberties critique of the EU AI Act centers on:
✓ Correct — Correct. Civil liberties critiques argue the Act does not protect people enough. Competitiveness concerns come from industry.
Civil liberties critiques focus on insufficient protections — surveillance exceptions and weak individual rights. Cost and competitiveness concerns come from industry, not civil liberties advocates.

Lab 4 — EU AI Act Critique

Evaluate the Act from a chosen stakeholder perspective

Your Task

Choose a stakeholder: civil liberties advocate, EU startup founder, large US tech company, or affected community group subject to algorithmic decisions.

Construct the strongest critique of the EU AI Act from that perspective — what does it fail to address, what does it get wrong, what would you change?

State your perspective and make your opening critique. I will push back and test your arguments.
AI Lab AssistantEU AI Act Policy Analyst
Choose your stakeholder perspective and give me your opening critique. I will challenge you to defend and sharpen your arguments.

Module Test

15 questions · 80% to pass
The EU AI Act applies to:
✓ Correct — Correct.
The Act is extraterritorial — it applies based on where people are affected, not where developers are based.
Government social scoring is classified as:
✓ Correct — Correct.
Government social scoring is in the prohibited tier — banned entirely.
High-risk AI systems must be designed so that:
✓ Correct — Correct.
Human oversight is a design requirement — capability for meaningful supervision, not per-output approval.
Both providers and deployers have obligations under the Act because:
✓ Correct — Correct.
Both have separate obligations. A deployer cannot shield itself by claiming it only purchased, not built, the system.
All GPAI model providers must publish:
✓ Correct — Correct.
All GPAI providers need training data summaries and usage policies. Full source code and complete datasets are not required.
The systemic risk threshold for GPAI models is:
✓ Correct — Correct.
The trigger is training compute — 10^25 FLOPs.
Systemic risk GPAI models must additionally:
✓ Correct — Correct.
Systemic risk models require adversarial testing, incident reporting, cybersecurity measures, and energy reporting.
For most high-risk AI systems, conformity assessment is conducted by:
✓ Correct — Correct.
Most high-risk systems use provider self-assessment. Third-party review is only required for specific sensitive categories like biometric identification.
Which AI is explicitly excluded from the EU AI Act?
✓ Correct — Correct.
Military and national security AI is explicitly out of scope.
The Brussels Effect describes:
✓ Correct — Correct.
The Brussels Effect is about regulatory diffusion — EU rules spreading globally through market behavior.
Prohibited practice rules became enforceable:
✓ Correct — Correct. Prohibited practices were first — establishing hard limits before the broader compliance framework activated.
Prohibited practices activated first — 6 months. GPAI at 12 months; most high-risk at 24 months.
A company buying and deploying a high-risk AI hiring tool without building it is classified as:
✓ Correct — Correct.
Organizations that use AI in their operations rather than building it are deployers, with their own compliance obligations.
EU AI Act penalties for the most serious violations reach:
✓ Correct — Correct.
Maximum penalties are €35 million or 7% of global annual turnover.
Civil liberties critiques of the EU AI Act focus on:
✓ Correct — Correct.
Civil liberties critiques argue the Act protects people insufficiently. Cost and competitiveness concerns come from industry.
The EU AI Act high-risk risk management system requirement is:
✓ Correct — Correct.
Risk management under the Act is continuous — maintained throughout the AI's operational lifespan, not a one-time event.