AI: The Risks That Matter — and the Rights You Already Have
Learning objective
After reading this article, you will know the five structural risks AI poses at the collective level, you will be able to tell what is established from what is debated. And above all, you will know the concrete rights European regulation already gives you when facing AI systems, because almost nobody has told you.
Why this article differs from the previous three
The AI track has equipped you for your personal use: the reflexes, how it works, deepfakes. But AI doesn't only concern you when you use it: it concerns you when others use it on you, to sort your job application, score your credit, analyze your face, fill your feed.
That is the collective level, and it has its own risks. This article lays them out honestly (including where specialists disagree), and does so with a practical bias: for each risk, what you can concretely do about it.
The 5 structural risks
Risk 1 — Bias, at scale
You know where biases come from: training data reflects human texts and decisions, imbalances included. The change of scale is the real issue: a biased recruiter affects the candidates they see; a biased sorting system, deployed everywhere, affects millions of applications, with an appearance of objectivity no human ever had. Documented cases exist in recruitment, credit scoring and facial recognition, whose error rates have varied significantly depending on the faces analyzed.
What you can do: for decisions that concern you, Europe already arms you: the GDPR gives you the right not to be subject to a fully automated decision with significant effects (credit refusal, hiring pre-selection…) without being able to obtain human intervention, express your view and contest the decision. "It's the algorithm" is not a legally sufficient answer: you can demand a human in the loop. (Outside the EU, protections vary, but the reflex of demanding human review is worth having everywhere.)
Risk 2 — Surveillance made cheap
Surveillance has always existed; AI changes its economics. Analyzing faces in a crowd, transcribing and cross-referencing conversations, inferring habits from traces: what once required entire teams becomes automatable. The structural risk is right there: what becomes cheap becomes tempting, for commercial and state actors alike.
Europe has drawn lines: the AI Act prohibits certain uses deemed unacceptable (generalized social scoring, certain forms of mass biometric identification), and places high-risk systems under reinforced obligations. The debate remains open on the right security/liberties balance, and diverging positions legitimately exist.
What you can do: at the individual level, your lever remains exposure reduction, less raw material, less possible analysis. And your GDPR rights (access, rectification, erasure) also apply to the data feeding these systems.
Risk 3 — Work transformed
Here, honesty requires laying out a debate rather than a conclusion. What is established: AI transforms tasks (writing, analysis, code, customer service) faster than previous technological waves, and shifts the value of skills. What is debated: the net balance. Some economists anticipate massive, rapid job destruction; others recall that every previous wave destroyed trades and created entire sectors, and foresee transformation rather than disappearance. Both camps include serious people, and current data doesn't settle it.
What you can do: in both scenarios, the same individual variable makes the difference, usage competence. Whoever knows how to put AI to work (delegate the first draft, keep the judgment, verify) turns the tool into leverage; whoever ignores it, endures it. This entire track is precisely that insurance.
Risk 4 — Polluted information
The cost of producing plausible text, images or video has fallen to nearly zero: the cost of verification has not. That asymmetry is the systemic risk: mass-generated content, targeted fakes (deepfakes), and a background noise that makes the authentic harder to distinguish.
What you can do: your reflexes already exist. verification before sharing, the context test. Add an ecosystem-level act: give your attention (and where relevant your money) to sources that verify before publishing. In an attention economy, that is a vote.
Risk 5 — Delegated judgment
At the individual level, you know mistake 5 from the foundations: letting AI think in your place. The collective risk is more serious: administrative, medical and judicial decisions increasingly assisted by systems. And the well-documented bias that comes with it, that of following the machine because it's the machine, even when it's wrong. The danger is not that AI decides: it's that the human supposed to oversee stops truly overseeing.
What you can do: at work, if AI tools take part in decisions, keep track of the why of each decision. "The tool suggested it" must never be the complete justification. And for decisions aimed at you, the right to human intervention from risk 1 exists exactly for this: use it.
What really matters
1. AI amplifies — it doesn't invent. Bias, surveillance and disinformation existed before it. What it changes is the scale and the cost. Understanding this avoids the two symmetrical errors: panic and denial.
2. You already have rights — use them. Human intervention on automated decisions (GDPR), mandatory chatbot transparency and synthetic-content labeling (AI Act), control over your data. These rights only have force when invoked.
3. Competence is the best individual protection. Against transformed work and polluted information alike, knowing how to use and evaluate AI is the insurance that works in every scenario.
A simple method to put in place
This week: take the inventory. Where does AI already decide for you or about you? Feed, recommendations, application sorting if you hire or apply, possible scoring. Naming the places is the first step.
This month: if an important automated decision has already been opposed to you (credit, application, administration), know that you can request human review. And do it if still relevant. Your national data protection authority documents these rights and takes complaints (France's CNIL, Spain's AEPD, Bulgaria's CPDP, the UK's ICO).
Ongoing habit: at work and at home, when an AI tool takes part in a decision, one ritual question: who owns this decision, and do they know why it was made? If nobody can answer, the human loop is broken, that's the signal.
What "good enough" looks like:
- Beginner: you know that important fully automated decisions can be contested, and you can tell what's established from what's debated about jobs
- Intermediate: you've inventoried the systems deciding about you, your exposure and verification reflexes are in place, and you know where to complain (your data protection authority)
- Advanced: in your own work, the human in the loop is real and documented, and you can explain both the risks and the rights to those around you
An honest note
This article lays out debates that specialists themselves haven't settled, on jobs, on the regulation/innovation balance, on the relative gravity of the risks. That's deliberate: giving you the established facts and the map of positions equips you better than a hammered conclusion. On these subjects, be symmetrically wary of apocalypse prophets and utopia salesmen: both spare you the trouble of thinking.
And keep the track's perspective: the individual answer to collective risks is neither fear nor ignorance, it's competence, plus rights. You now have both.