---
title: "The EU AI Act (video series)"
description: "A five-part series on the EU AI Act (Regulation 2024/1689): why a regulation on AI reaches your HR team wherever you are based, the four-tier risk pyramid, why hiring and performance AI is treated as high-risk even when you only bought the tool, what is enforced now versus what the 2026 Digital Omnibus pushed to December 2027, and the hard part, proving a human was genuinely in charge. Hosted by Deepti Gupta, co-founder of Sageo. General information, not legal advice."
url: "https://sageo.ai/videos/eu-ai-act"
updated: "2026-11-09T00:00:00.000Z"
source: "Sageo (sageo.ai)"
---


# The EU AI Act

_A five-part series on the EU AI Act (Regulation 2024/1689): why a regulation on AI reaches your HR team wherever you are based, the four-tier risk pyramid, why hiring and performance AI is treated as high-risk even when you only bought the tool, what is enforced now versus what the 2026 Digital Omnibus pushed to December 2027, and the hard part, proving a human was genuinely in charge. Hosted by Deepti Gupta, co-founder of Sageo. General information, not legal advice._

Full playlist: [The EU AI Act on YouTube](https://www.youtube.com/playlist?list=PLaKugtwEK7i0)

## Part 1: The AI Law That Reaches Your HR Team

The EU AI Act (Regulation 2024/1689) is the first broad law anywhere to regulate AI across every sector, and it has been in force since August 2024. Because it is a regulation and not a directive, it applies directly and identically in every EU country, with no national versions to wait for. And because it regulates how AI is used rather than the technology itself, the same model is minimal risk when it drafts marketing copy and high-risk when it scores job applicants. What brings it to your team is where the AI is used, not where your company is based, so evaluating staff in France or screening candidates in Germany is in scope whether you sit in London, Bengaluru, or San Francisco. Three things carry across all of it: know where AI is used, keep a human able to oversee it, and document the decisions.

Watch: [The AI Law That Reaches Your HR Team](https://www.youtube.com/watch?v=1DfJS9EMoU0)

## Part 2: The AI Risk Pyramid in Two Minutes

The Act sorts every use of AI into four tiers of risk. Unacceptable-risk uses are banned outright, and have been since February 2025: social scoring, scraping faces from the internet to build recognition databases, manipulation, and emotion recognition in workplaces and schools, the one most people miss. High-risk is where almost all the real obligations sit, covering AI built into regulated products like medical devices and vehicles, and named areas including credit scoring, education, critical infrastructure, law enforcement, and employment. Limited-risk uses carry transparency duties, so a chatbot has to say it is a machine and AI-generated content is labelled. Everything else is minimal risk, with no specific obligations. Most AI is minimal, but the obligations pile up in the high-risk tier.

Watch: [The AI Risk Pyramid in Two Minutes](https://www.youtube.com/watch?v=SXHTFbbuL_Q)

## Part 3: Why Your Hiring and Performance AI Is High-Risk

Using AI to screen CVs, score interviews, rate performance, or monitor workers sits squarely in the high-risk tier, even if you bought the tool rather than built it. The Act names the employment uses directly: hiring, promotion, termination, task allocation, and performance monitoring. The distinction that matters most is provider versus deployer. A provider builds the system and puts it on the market; a deployer uses it under their own authority, and most companies are deployers (build your own internal tool and you are both). Deployers carry real duties: use the system the way the instructions require, make sure human oversight actually happens, monitor it and keep the logs, feed it relevant and representative data, tell affected workers, and for some uses run a fundamental rights impact assessment. Human oversight is the heart of it: a real person who understands what the system recommended, can overrule it, and leaves a record of the call.

Watch: [Why Your Hiring and Performance AI Is High-Risk](https://www.youtube.com/watch?v=656QQw89_6c)

## Part 4: The EU AI Act Timeline: What's Live Now and What's Coming

What is enforced now, and what is still coming. The prohibited practices have applied since February 2025, and obligations for general-purpose AI models since August 2025. The big date for people teams is 2 December 2027, when employment AI becomes high-risk and hiring, performance, and promotion systems face the Act's strictest duties: human oversight, transparency, and logging. That date moved: the 2026 Digital Omnibus pushed the high-risk deadlines back, but it moved the clock without changing the direction, so the high-risk rules are still coming. The penalties are sized to make the point. Up to 35 million euros or 7 percent of global annual turnover for banned practices, and up to 15 million euros or 3 percent for most other breaches including high-risk duties, whichever is higher, and for startups and SMEs whichever is lower.

Watch: [The EU AI Act Timeline: What's Live Now and What's Coming](https://www.youtube.com/watch?v=GGizwJQmwI0)

## Part 5: The Hard Part: Provable Human Oversight

Most of the Act can be handled with documentation. The hard part is human oversight, and real oversight is not a person clicking approve. It means someone who can understand the AI's recommendation, see why it was made, overrule it, and record that they did, with the decision tied to clear, documented criteria your managers already use. If those criteria are vague, the AI amplifies the vagueness; if they are clear, most of the oversight problem is already solved. The series closes with a six-step readiness plan: inventory your AI, classify each use, work out whether you are a provider or a deployer, set up human oversight (the part that takes longest, so start there), fix the criteria underneath with clear levels and competencies, and sort out transparency and records. Sageo builds the human oversight and evidence layer, competency frameworks and an auditable record of manager judgement, while your lawyers handle legal classification and your vendors answer for their own models. 2 December 2027 sounds far away, but oversight takes longer to build than you think.

Watch: [The Hard Part: Provable Human Oversight](https://www.youtube.com/watch?v=wh4rinfQ1Rw)
