Remember Cambridge Analytica? Most people treated their online activity as harmless. You clicked a quiz, liked a post, sent an application, and moved on. Later, we learned that data on tens of millions of people had been quietly harvested, combined into detailed voter profiles, and fed into political campaigns in several countries, including Canada. Ordinary behaviour had been turned into a hidden scoring system that helped decide what you saw and what you never even knew was there.
Hiring tech is building its own version of that scorecard.
AI hiring tools are sold as simple helpers. Quicker screening, better matches, less bias. They plug into your ATS, sit inside your existing workflow, and start learning from everything that moves through your pipeline. Over time, they are in a perfect position to turn candidate behaviour into scores that decide who gets surfaced, who gets buried, and who never makes it past the first filter.
The Eightfold lawsuit is an early look behind the curtain. The people suing are not long-shot applicants. They are seasoned STEM professionals with more than ten years’ experience who applied to hundreds of roles at employers using Eightfold’s tools and watched their applications stall with no clear reason. At some point, it is natural to wonder whether your résumé is still the main thing making the decision.
According to the complaint, it is not. They allege that Eightfold builds profiles and scores on jobseekers using their applications plus external signals, including third-party data and online activity, then feeds those scores to employers for screening and ranking. Candidates are not clearly told that a score exists; they never see what’s in it, and they have no way to correct it, even though that score may be deciding whether they move forward.
The plaintiffs argue that if a third party is assembling dossiers and “Match Scores” on workers and selling them into hiring decisions, it starts to look like a consumer reporting agency. In that view, those AI-generated profiles are consumer reports in everything but name and should trigger the same basic rights we expect from a credit bureau: notice, access, and the ability to dispute errors.
Eightfold disagrees. It says it uses data from candidates and clients, complies with privacy and employment law, and exists to help employers make better, less biased decisions. It rejects the idea that it is acting as a credit bureau for workers and describes itself as a tool that makes hiring more efficient, not a hidden scoring agency. The legal line will be drawn in court.
The bigger story is the structure we are building around hiring.
People will say, “But it’s only one system. Not every company uses the same platform. How could it possibly build real profiles on the same person across employers?” The thing is, this is exactly how data power has worked in other markets. You do not need every lender to use the same bureau or every site to use the same ad network. You just need a handful of big intermediaries in the middle, plugged into enough players, to effectively follow people around.
Hiring platforms live in that middle seat. They integrate with many employers, not just one. They can often link activities using the same email address, name, work history, LinkedIn profile, or other details that appear again and again when someone applies. On top of that, many contracts let vendors use “aggregated” or “de-identified” data to improve models for all customers. In plain terms, the system is allowed to learn from what happens to people like you across its client base, even if it does not show your name back to other employers.
A “profile” in this context is not a tidy file with your name on it that HR can open. It is a learned pattern, a statistical picture of you, or of people the system decides look like you on paper and online. That pattern can include things like skills, education, experience, location, past application behaviour, how you respond to offers, and typical pay levels. When you reappear in the system, that pattern is already in the background, shaping how you are scored and how visible you are to employers.
In Canada, we have privacy legislation, human rights protections, and early guidance for AI in employment. On paper, we talk about transparency, human-in-the-loop, and limits on fully automated decisions. In practice, those rules assume you can point to a visible decision and understand it. They are not designed for a system that sits in the background, quietly adjusting your rank based on information no one outside the platform can see. Our legal system also moves after the fact. It steps in when someone can demonstrate harm or a clear violation. That creates a gap between “we have a law” and “we can enforce it inside real systems, at scale.”
In consumer markets, small steps add up. Collect more data here, personalize a bit more there. Over time, that became algorithmic price discrimination. Once companies could estimate how much different customer types would be willing to pay, they no longer had to charge everyone the same price. They could quietly charge more to those who were likely to accept it. No one needed a big decision labelled “discriminate.” The system simply optimized the information it had.
In the gig economy, companies like Uber and Amazon have already shown how this works in practice. Their systems watch when people log in, which trips or tasks they accept, when they walk away, and how they react to pay cuts or bonuses. Over time, they can infer each worker’s “floor,” then tune offers so that two people can be paid differently for essentially the same job, simply because the model has learned that one will tolerate less.
The Eightfold case does not claim this is already happening in corporate hiring. It describes something that comes just before that. A hiring platform that allegedly builds cross-employer dossiers and scores on workers, feeds them into decisions, and leaves the people being scored in the dark is building exactly the kind of infrastructure you would need if you wanted, later, to bring the same optimization logic to who gets seen and what they are offered.
Think about the view an AI system gets when it is wired into an ATS over time. It sees which roles people apply for, where they are screened out, who gets offers, who accepts, and, often, the compensation attached to those offers. From that, a model can start to learn which profiles usually say yes to which kinds of roles, at which salary levels. Once you have that, you can start adjusting both visibility and pay to match those patterns.
You do not need a bad actor for this. You just need a clear objective. If the objective is to fill roles quickly at the lowest sustainable cost, and the system can see how different candidate profiles respond to different offers, it will naturally move toward offering as little as it can while still getting a yes. That is how optimization works. You can have well-intentioned people overseeing the system and still end up with outcomes that tilt bargaining power further away from workers.
This isn’t just a candidate problem. It affects employers as well. When you buy a tool like this, you are not only buying speed. You are buying a particular view of the labour market, shaped by the platform’s data and incentives. You can end up locked into its definition of “qualified,” its idea of “acceptable pay,” and its quiet decisions about who is worth surfacing. On the surface, it appears to be efficient. Underneath, you may be narrowing your pipeline and hardening pay practices you would never sign off on if they were described plainly.
The platform sits in the strongest position in this relationship. On paper, employers own their data. In reality, the vendor owns the logic on what gets logged, how models are trained, and how insights get pushed back into your hiring. It sees many companies over long periods. It knows who is often rejected, who tends to accept quickly, and at what pay level. That information is valuable. Markets rarely leave valuable information unused for long.
So, the question is not whether one vendor is secretly malicious. The more useful question is what kind of system are we building when we hand detailed hiring data to platforms with clear commercial incentives and very little external visibility? In other domains, when platforms learn about buyers, they push prices toward each buyer’s limit; when they learn about workers, they push wages toward each worker’s limit. If hiring platforms now learn about applications, outcomes, and pay, it is reasonable to expect that access and offers will be pushed toward whatever best serves their customers and their own business.
It probably will not be marketed in those terms, and it will not live in a settings menu. Systems do not need a label for a behaviour before they develop it. They need data, a clear objective, and time.
The Eightfold case might fail on legal technicalities, or it might succeed and force changes. Either way, it is a warning shot. The long-term story of AI in hiring is unlikely to be written by candidates, and not fully by individual employers. It is being written, quietly, by the platforms in the middle. If you are buying these tools, that is the story you are stepping into. The responsible move is not to panic, but to ask better questions, insist on real transparency, and make sure you understand whether you are getting help or handing your hiring strategy to a hidden scorecard that sees more of your candidates’ lives than you do.