Est.

Legal Considerations When Sourcing Candidates From Public Web Data

Scraping candidate data from public sites triggers multiple legal regimes at once.

· 10 min read
Cover illustration for “Legal Considerations When Sourcing Candidates From Public Web Data”
Talent Intelligence · October 13, 2026 · 10 min read · 2,336 words

Picture a recruiter who builds a sourcing tool that pulls candidate names off professional networking sites, code-hosting platforms, and a dozen conference speaker pages, then feeds that into an AI model that ranks a thousand people before lunch. That used to be a side project for a scrappy startup. Now it's the default setup at companies of every size. The legal frameworks built for weird edge cases now sit on top of everyday recruiting work.

Sourcing candidates from public web data is now a compliance-critical activity

A sourcing tool queries public profiles and builds a ranked candidate list, so it isn't an experiment anymore. Source a remote candidate in Illinois or Colorado, and that state's rules travel with the data. A company's AI strategy matters less than whether its data and decisions meet these legal obligations. A name attached to a job title is a legal object now, not just a lead.

A candidate's profile sitting on a public web page doesn't exempt it from legal scrutiny. What matters is how that data got collected and what happened to it next. There's no single law that governs web scraping worldwide. Instead, several frameworks apply at once: the CFAA in the US, GDPR in the EU, CCPA in California, copyright law, and the terms of service on whatever platform hosts the profile. LinkedIn, decided in 2019 and again in 2022, held that scraping publicly available profiles likely doesn't violate the CFAA, because "without authorization" in that statute targets breaking through access controls, not ignoring a platform's stated preferences about public pages.

That ruling didn't end the hiQ saga, though. The case wrapped up in a December 2022 consent judgment and private settlement, with a permanent injunction, deletion of all scraped data, and a $500,000 payment to LinkedIn. A more recent case sharpens the same lesson. LinkedIn Corp. v. ProAPIs Inc. ended in a final consent judgment signed September 16, 2026, by Judge P. ProAPIs had been selling real-time access to LinkedIn member data through an API that handled up to 150 requests per second, at subscription prices that reached $15,000 a month, a business model that got dismantled in federal court.

The durable rule out of this run of litigation: scraping public pages from the open web is mostly a contract question, governed by the platform's terms of service. For a sourcing pipeline, legal exposure turns on the collection method (direct open-web queries versus authenticated session scraping), not simply on whether the destination page happened to be publicly visible. That distinction sets up everything that follows, because clean access under contract law buys exactly zero protection under privacy law.

How GDPR and CCPA reach publicly posted candidate data

Privacy law doesn't care whether the collection method was legal. It also imposes separate obligations on what happens to the data once you have it, and publicly posted information about an identifiable person is still personal data under GDPR and personal information under CCPA. GDPR also requires a lawful basis for processing that data, and the basis most often invoked for web-sourced candidate data is "legitimate interest" under Article 6(1)(f). That basis comes with a proportionality balancing test: the company's interest in collecting the data can't override the individual's own rights.

The Clearview AI enforcement sequence is the clearest signal of how regulators treat the "it's public, so it's fine" defense. Clearview argued that its scraped photos were publicly available and that it had a legitimate interest basis for using them. EU regulators, led by a national data protection authority, rejected that argument. Regulators weigh it against a basic expectation that a public profile won't get vacuumed into an employment scoring system without the person's knowledge.

California has moved on a parallel track. Initial assessments covering processing that was already underway before 2026 are due by December 31, 2027, with the first submission to the CPPA due April 1, 2028. Retention has a real clock on it, too. Past that window, the data needs to be anonymized or deleted unless the candidate specifically agreed to longer retention for future consideration.

Anti-discrimination law and liability for automated candidate selection

Collecting the data lawfully doesn't close the file. The moment an algorithm starts ranking or filtering candidates, a second, older body of law kicks back in. Title VII, the ADA, and the ADEA apply to AI-driven candidate selection on the same disparate impact theory they've always applied to human decisions, and both the employer and the vendor supplying the tool can end up on the hook. There's no federal statute written specifically for AI in hiring, but Title VII, the ADA, and the ADEA impose liability for discriminatory outcomes no matter whether a person or an algorithm produced them.

Vendor liability got a concrete test in Mobley v. Workday, Inc. A plaintiff alleged that Workday's resume-screening AI disproportionately excluded older applicants and other protected groups. The plaintiff also alleged that Workday trained its AI on the employer's existing workforce data, which caused the system to replicate the employer's own biases around age, race, and disability. So training data choices become legal exposure down the line, not just a modeling decision. On the employer side, the stakes are already measured in dollars: less than a year before the Mobley ruling, a virtual tutoring company paid $365,000 to settle EEOC claims that its AI recruitment tool screened out candidates based on age.

The "black box" nature of a lot of these systems makes the exposure worse. California's regulations build in one way out: anti-bias testing and corrective action, done proactively, count as an affirmative defense against disparate impact claims. So that fact turns compliance from a defensive posture into something closer to insurance.

The emerging FCRA theory: when AI candidate profiling becomes consumer reporting

A newer legal theory is working its way through the courts, and it has nothing to do with bias. Class action plaintiffs are arguing that AI platforms scraping, scoring, and ranking candidates at scale are functioning as unregistered consumer reporting agencies under the Fair Credit Reporting Act. If that theory holds, it stacks an entirely separate compliance regime on top of everything already covered here.

In January 2026, plaintiffs Erin Kistler and Sruti Bhaumik filed a class action against Eightfold AI Inc., alleging its AI hiring platform operated as an unregistered consumer reporting agency under the FCRA. Unlike most AI hiring lawsuits, which argue bias, this one argues privacy and disclosure failure: that Eightfold compiled and used personal data without adequate disclosures, consent, or any way for a candidate to dispute what had been collected about them.

This theory didn't appear out of nowhere. The CFPB issued guidance stating that algorithmic employment scores qualify as FCRA-covered consumer reports. If the theory succeeds in court, the practical requirements that follow would include disclosures to candidates before a report gets used, adverse action notices when a low score leads to rejection, and a formal dispute process, obligations that most current sourcing pipelines simply aren't built to handle.

The state legislative wave now in effect and taking effect through 2027

States have stopped waiting for the federal government. On July 1, 2025, the US Senate voted 99-1 to strike a moratorium on state AI laws out of the One Big Beautiful Bill Act, and the version President Trump signed contains no such moratorium. That vote settled the question of federal preemption: state legislatures are free to keep building, and they have been.

Several laws are already live. New York City's Local Law 144, effective since 2023, bars employers and employment agencies from using automated employment decision tools for hiring or promotion in the city unless the tool has gone through an independent bias audit within the past year, a public summary of that audit sits on the employer's website, and candidates get at least 10 business days' notice before the tool is used on them. Illinois passed its Artificial Intelligence Video Interview Act back in 2020, and it requires candidate consent before AI analyzes a video interview. California's Civil Rights Council regulations took effect October 1, 2025, and they explicitly prohibit California employers from using AI or automated decision systems that harm applicants based on protected characteristics like gender, race, or disability, with unintentional disparate impact still creating liability (though anti-bias testing can serve as a defense). Illinois followed up with amendments to its Human Rights Act, effective January 1, 2026, requiring employers to notify applicants and employees that AI will be used in hiring and recruitment decisions, and making discrimination through AI tools explicitly illegal. California's Privacy Risk Assessment rules, also effective January 1, 2026, apply whenever ADMT drives a significant decision about a consumer, including an employment decision.

More is coming through 2027. Colorado originally planned to bring its Artificial Intelligence Act into force on February 1, 2026, but the state repealed that version and replaced it with a new ADMT law set to take effect January 1, 2027. California's ADMT rules under the CCPA, finalized in the fall of 2025, take effect January 1, 2027, applying to CCPA-covered businesses using ADMT for significant decisions, including employment, and requiring pre-use notice along with opt-out rights. Connecticut passed broad legislation covering what it calls automated employment-related processes, or AEDPs, requiring employers to disclose in plain language that a candidate is interacting with one of these systems, explain its purpose and the nature of the decision it makes, and offer an opt-out before it can affect an employment outcome; the employment provisions take effect October 1, 2027. California also requires, starting in January 2026, that vendors of generative AI tools publicly post information about the data used to train them.

A pattern runs through all of these statutes: notice to candidates, a chance for a human to review adverse automated decisions, and accountability through audits and impact assessments on a fixed schedule. A team sourcing candidates across state lines can't treat the toughest applicable law as a special case. That law is the operational floor for the whole pipeline.

The EU AI Act in a global sourcing pipeline

Organizations sourcing candidates in or from the EU carry an extra layer on top of everything above. The EU AI Act came into force August 1, 2024, and it classifies AI systems used in hiring and employment as high-risk under Annex III, which triggers transparency requirements, human oversight obligations, and conformity assessments. The EU's Digital Omnibus entered into force on July 27, 2026, and it pushed back the Annex III high-risk obligations, originally due to apply from August 2, 2026, to December 2, 2027.

That delay buys time, but it doesn't erase the obligation, and it shouldn't be read as permission to sit still. Organizations without compliant controls in place should treat the extra runway as time to build them, not time to wait around. GDPR, notably, isn't touched by any of this. Put the two frameworks together: an EU-facing sourcing pipeline ends up with GDPR governing the data layer and the AI Act governing the automated decision-making layer on top of it. That produces overlapping audits and documentation requirements that a compliance program built only for US law has no chance of satisfying.

Building a sourcing pipeline that satisfies this layered compliance map

No single checkbox covers all of this. A defensible sourcing pipeline comes from design choices made across every layer at once: access law, privacy law, anti-discrimination obligations, and the state statutes now stacking up through 2027.

Start with how data gets collected. Keep the data structured and clean rather than raw HTML, so every data point feeding the AI model is auditable and traceable back to its source; unstructured scrapes make bias audits and data subject access requests nearly impossible to satisfy later.

On minimization and retention, collect only the fields you actually need for the stated recruitment purpose. Set a default retention window of six to twelve months for unhired candidates, with automatic deletion or anonymization at expiry unless a candidate has explicitly agreed to a longer hold.

On transparency, give candidates notice that their data was sourced from the web. That's already a GDPR requirement, and it will be mandatory under California's ADMT rules starting January 1, 2027, and under Connecticut's law starting October 1, 2027. Illinois, effective January 1, 2026, and Connecticut both require disclosure before an AI tool is used in the employment process at all, so that disclosure belongs in the candidate-facing interface, not buried three pages into a privacy policy. Build a working process for data subject access, correction, and deletion requests now, since retrofitting one under enforcement pressure costs a lot more than building it in from the start.

On bias audits and oversight, test every AI tool in the pipeline, including anything bought from a vendor, for disparate impact before it goes live and on whatever schedule the applicable law sets: annually under NYC Local Law 144, within 90 days of material changes under the original Colorado framework, and on whatever cadence Colorado's new implementing rules eventually specify. Train HR teams and hiring managers on what these tools actually do, where they fail, and what the law requires of them. That's a distinct compliance step, not a feel-good training session.

On vendor contracts, get written commitments on data sources, training data composition, bias testing, and FCRA compliance posture. Mobley made clear that employer liability doesn't vanish just because a vendor built the tool. Confirm, in writing, that the vendor's own data collection practices comply with the terms of service of the platforms they're pulling from.

On the FCRA frontier, assess whether any scoring-and-filtering tool in the pipeline is functioning as a consumer reporting agency before a human ever reviews the candidate. If there's any plausible argument that it is, build the adverse action notices and the dispute process now, before a court forces the issue. The Eightfold AI litigation is the one to watch. Its outcome will decide whether this becomes a permanent fixture of hiring compliance or stays a contained legal skirmish.

Sources

  1. Efficiency to Exposure: The Global Compliance Challenge of AI Recruiting
  2. With Federal Restrictions Removed, A Wave of State Laws Highlights Risks of Using Artificial Intelligence in Hiring and Employment Decisions: Stinson LLP Law Firm
  3. AI in Hiring: Evolving Legal Risks Under State and Federal Law: Wiley
  4. The Ever-Changing Legal Landscape of State and Federal Regulations for Using AI in Candidate Recruiting and Screening - Steptoe & Johnson PLLC
  5. AI-Based Hiring: 2026 Developments Employers Can’t Ignore

More in Talent Intelligence