Consent in the age of AI is mostly theater. And while individual rights remain important, they are nowhere near enough.
Why consent breaks down
The Fair Information Practices (FIPs) — the backbone of modern privacy law — were designed in the 1970s. They include:
- Notice and choice
- Purpose limitation
- Data minimization
- Access and correction
- Use limitation and safeguards
These principles assume one thing: the individual is in control.
But that assumption collapses in today’s AI landscape. AI systems process massive volumes of data — much of it indirect, inferred, or repurposed. You often don’t know it’s happening, let alone have the tools to stop it.
Take a simple scenario: you apply for a job. Your CV is analyzed by an AI model. That same data is later reused to improve a separate productivity tool — without your knowledge.
That violates purpose limitation. But how would you opt out? You probably didn’t even know it happened. And even if you did, would you really have the time or expertise to navigate legalistic privacy notices?
The reality: we click “accept” not because we understand — but because we have no alternative. This is why Daniel Solove calls the privacy paradox a myth. People do care about privacy. They’re just set up to fail.
How AI makes it worse
AI systems amplify the gap between legal theory and lived reality:
- They infer private traits you never disclosed (violating purpose limitation).
- They retrain on your data without notice (violating use limitation).
- They rely on opaque models you can’t interrogate (undermining access and correction rights).
Consent boxes can’t fix these systemic problems — and this is why it’s not necessarily a good direction that GDPR simplification (Omnibus IV) is heading. I understand the need to ease administrative burdens for SMEs, but in the age of AI, privacy risks are not tied to company size. Even a small start-up can process enormous amounts of sensitive data or make powerful inferences. Loosening safeguards based on headcount risks undermining the very protections AI makes more necessary.
Privacy by design and default: what it really means
So what’s the alternative? A shift from individual responsibility to organizational accountability — built on privacy by design and default.
Here’s what that looks like in practice, especially in HR and recruitment contexts:
- Minimization first: collect only what is necessary to evaluate a candidate’s suitability. Example: if the role doesn’t require travel, don’t collect detailed location histories or mobility data.
- Purpose clarity: candidate data given for job screening must be used for that purpose only. Example: CVs and assessment scores cannot later be repurposed for marketing campaigns or training unrelated productivity tools.
- Embedded safeguards: build technical protections directly into recruitment platforms. Example: use role-based access so only recruiters (not marketing teams or engineers) can see applicant data.
- Default protections: privacy settings should protect candidates automatically. Example: job portals should not default to “share profile publicly” — candidates should actively opt in.
- Transparency that informs: give candidates clear, plain-language explanations of how AI will be used in their evaluation. Example: instead of legal jargon, a short notice: “This AI tool helps screen CVs for skills. It does not make final hiring decisions. A recruiter reviews all rejections.”
- Accountability across the chain: require not just the vendor, but also the employer using the system, to prove compliance. Example: employers must demonstrate they review AI-driven shortlists for bias before acting on them.
A special note on bias audits: in some cases, ensuring fairness means collecting special category data (such as race, health, or age) to test for discrimination. This may feel at odds with data minimization requirements, but it is necessary to detect and correct bias. The EU AI Act explicitly allows such use for bias monitoring and correction — however, GDPR exceptions are less clear. Organizations should therefore apply strong safeguards (e.g., pseudonymization, strict access controls), and carefully document their justification for collecting and using this data.
Privacy by design and default is about building trust into recruitment workflows — ensuring candidates are not unknowingly profiled, repurposed, or excluded by opaque systems.
The shift we need
Consent still matters. But it must be informed, voluntary, and manifest — not buried in fine print.
That means moving away from privacy as a user burden and making it an organizational duty. Stronger data governance, enforceable purpose limitation, plain-language notices, and accountability across providers and deployers are the real safeguards.
Privacy by design needs an owner. Our fractional DPO builds and runs it.