The €15M Question

EU AI Act high-risk obligations enforce on 2 August 2026.

Non-compliance: up to €15,000,000 or 3% of worldwide annual turnover, whichever is higher.

• First comprehensive AI regulation anywhere in the world
• Less than 3 months away as of writing
• Most enterprise agents fall into "high-risk" — that's where the obligations bite

What this covers.

• The 7 articles that drive high-risk operational requirements
• GPAI obligations and what they mean for downstream builders
• Deployment topology and EU data residency
• A 90-day plan to get audit-ready

No runnable code. Conceptual diagrams and crosswalks only.

Risk Classification

Four tiers + a GPAI overlay.

Are you high-risk?

You're high-risk if your agent makes or materially supports decisions in:

• Financial services — credit scoring, insurance pricing
• Employment — recruitment, screening, performance evaluation
• Education — admissions, exam scoring
• Healthcare — medical devices, triage, diagnostic support
• Biometric identification — including emotion recognition
• Critical infrastructure, law enforcement, migration, justice

If you're unsure, assume high-risk and downgrade with legal counsel.

Quick reality check.

The Seven Articles

Seven articles. One crosswalk.

Articles 9, 10, 12

Foundation: Risk, Data, Logging

Art. 9 — Risk Management

A living risk management system across the development lifecycle — not a one-time document.

Requires: identify, analyze, evaluate, and mitigate risks — continuously updated.

LangChain v1: - Online evaluators scoring production traffic against custom thresholds - Custom evaluators for domain-specific risks — financial accuracy, clinical safety - Webhook → PagerDuty alerts when thresholds breach - Risk register kept in sync with evaluator outputs

Art. 10 — Data Governance & Bias

Data quality, representativeness, and explicit bias examination across protected characteristics.

Requires: documented data provenance, bias examination across race, gender, age, religion, nationality, disability, sexual orientation — and documented mitigations.

LangChain v1: - Bias and fairness evaluators — LangSmith ships templates per protected characteristic - PII Middleware — prevents leakage of protected attributes in inputs and outputs - Trace dataset documentation for evaluation provenance

Art. 12 — Automatic Event Logging

Logs spanning the full system lifecycle — sufficient for deployer oversight and regulatory inspection.

Requires: inputs, outputs, timestamps, agent context, sufficient detail for audit.

• EU residency — LangSmith EU SaaS, BYOC, or self-hosted

Articles 13, 14, 15, 72

Transparency, Oversight, Accuracy, Monitoring

Art. 13 — Transparency to Deployers

Outputs must be interpretable enough that deployers can use the system appropriately.

Requires: clear instructions, documented capabilities and limitations, interpretable outputs.

LangChain v1: - LangSmith Studio — visual execution graph showing state transitions, tool calls, decisions - Full reasoning traces — every step inspectable - Documented agent specs — inputs, outputs, tool registry, system prompt

Art. 14 — Human Oversight

Humans must understand, intervene on, override, and interrupt the system. Not theatrical — measurable.

Requires: oversight designed into the architecture, humans able to intervene at decision points, auditable trail.

Art. 15 — Accuracy & Adversarial Resilience

Declared accuracy levels and demonstrable protection against common attack surfaces.

Requires: stated accuracy metrics, adversarial resilience, consistency over system lifetime.

LangChain v1: - Correctness, exact match, plan adherence, task completion evaluators - Prompt injection and jailbreaking evaluators — LangSmith templates - API leakage, code injection evaluators for tool-calling agents - Adversarial evaluation suites — run before every release

Art. 72 — Post-Market Monitoring

Continuous monitoring of production behavior with incident reporting to authorities.

Requires: continuous monitoring, drift detection, incident reporting to national supervisory authorities.

LangChain v1: - Online evaluators with custom thresholds - Drift detection dashboards - Webhooks → incident response system - Audit dashboards for compliance and regulator-facing reporting

GPAI & Data Residency

GPAI — Provider or downstream developer?

Most LangChain users are downstream developers — not GPAI providers. Fine-tuning and redistributing can shift you into the provider role. Get legal advice if you're close to that line.

Deployment topology.

If you're in scope and your customers are EU-based, default to LangSmith EU SaaS or BYOC. The audit story is dramatically simpler when traces never leave the jurisdiction.

90-Day Compliance Plan

Days 1–30: Map.

• Identify which agents are high-risk under Article 6 + Annex III
• Document use case, deployer, sector, decision impact for each
• Determine your role: GPAI provider or downstream developer
• Stand up the risk register
• Pick deployment topology: EU SaaS / BYOC / self-hosted

Days 31–60: Wire up.

• LangSmith tracing on every high-risk agent path
• PIIMiddleware on inputs and outputs
• Bias evaluators per relevant protected characteristic
• LangGraph interrupt on every state-changing tool call
• Online evaluators for prompt injection and adversarial inputs
• Webhooks → incident response system

Days 61–90: Document.

• Technical documentation per Annex IV
• Risk management documentation per Article 9
• Logging and retention policies per Article 12
• Human oversight procedures per Article 14
• Post-market monitoring plan per Article 72
• Internal audit-readiness review before the August deadline

Documentation burden is real — budget engineering time, not just legal time.