Elusive Thoughts: A recruiter, a hacker and data broker you never spoke

TL;DRRecruiting tools and data brokers draw on one shared pool of contact data, and every leak from that pool hands scammers a phone number with a job title attached. I found documented cases for each link in the chain. I found no source tying a named scam wave to a named breach. "AI watermarking" in hiring turned out to mean hidden text in CVs, and nothing that traces leaks. Mint a unique alias per signup and ring recruiters back on a number you looked up yourself. Companies should treat staff phone numbers as attack surface.

A recruiter you never spoke to can hold a score on you and your mobile number. Then a broker resells the number and a criminal steals the database. A stranger rings you, pretending to be that recruiter.

Your phone number is a sales lead first and an attack surface second, and both lives run on the same data.

What AI sourcing actually does

It pulls public profiles, adds an email and a phone number, ranks people against a job, then sends the message or places the call itself. Each step has a legal fight attached.

In January 2026 two job seekers filed a proposed class action against Eightfold AI in Contra Costa County Superior Court in California. They allege it acts as a consumer reporting agency under the Fair Credit Reporting Act, scoring candidates from zero to five using data they never supplied, with no way to see or dispute it. Eightfold told Fortune it does not scrape social media and that the claims lack merit. These are allegations. I found no ruling.

The contact enrichment side already has a fine. In December 2024 the French regulator CNIL fined KASPR €240,000. Its Chrome extension revealed contact details for the LinkedIn profiles customers visited, from a database of about 160 million contacts. CNIL found it had collected data from people who had limited their visibility to connections and kept it too long. It also told people only from 2022, roughly four years after launch. That is a GDPR Article 14 failure with a price on it.

Scraping supply is a fight too. LinkedIn alleged ProAPIs ran millions of fake accounts, with access rented for up to $15,000 a month. In September 2026 a federal judge finalised a consent deal: the firms stop scraping and selling, and delete what they hold. ProAPIs says it never offered scraping tools.

The ICO audited AI recruitment tools in 2024 and made nearly 300 recommendations. It found tools gathering far more than needed and keeping it indefinitely, without candidates knowing. On the calling side, Phenom sells a voice agent that phones applicants and scores answers from 1 to 5.

What "AI watermarking" really means here

Four readings, one with solid evidence.

Hidden text in CVs. This is real. Greenhouse told the New York Times about 1% of the résumés it reviewed in early 2025 carried a trick, and ManpowerGroup said it finds hidden text in roughly 100,000 a year. A USENIX Security 2026 study of 196,682 résumés from hireEZ also found about 1%, and over 90% of those were stuffed keywords. Few were instructions to the model. Nobody measured whether it works. Counter-measures are blunt: auto-reject, or recolour the font and phone the candidate.

Detecting AI-written CVs. Google's SynthID Text is a genuine watermark, but it only flags text from models that embed it and weakens after rewriting or translation. I found no evidence recruiters check CVs for watermarks. Vendors market AI detectors, and I could not verify their accuracy.

Watermarking candidate data to trace leaks. I could not find credible sources for this practice. But I curious, does it happen???

Invisible identifiers in outreach. Tracking pixels are real. France's CNIL adopted a recommendation on 12 March 2026 saying pixels measuring opens generally need prior opt-in when the recipient never asked for the email. Cold outreach is exactly that case. I did not verify the ICO position.

Same pool, different shopfronts

The 2025 exposure shows the overlap. Researchers Bob Diachenko and nexos.ai found an open 16TB database of about 4.3 billion professional records. Contact details and work histories sat alongside enrichment metrics and Apollo IDs. It sat open from 23 to 25 November 2025. The owner was never confirmed.

Breaches I could confirm:

National Public Data: breached in late 2023, sold in 2024. Have I Been Pwned lists about 134 million unique emails, with phone numbers and addresses, though it flags the dataset as unverified. Reports put the haul near 2.9 billion rows. The company first said 1.3 million people, then conceded "hundreds of millions" and filed Chapter 11 on 2 October 2024.
DemandScience, a B2B demand-generation firm: about 122 million unique accounts, business emails and job titles, with business phone numbers per ITPro. The company blamed a decommissioned system.
Aura, March 2026: about 900,000 marketing contacts from a company it acquired, taken after a vishing call to an employee. ShinyHunters claimed it. Aura is not a broker, and I could not confirm a 2026 breach of a classic people-search site. Have I Been Pwned found about 90% of those emails already sat in earlier breaches.

The regulatory fix exists only in California. Californians could file with DROP from 1 January 2026, and over 600 registered brokers had to start processing requests on 1 August 2026, checking at least every 45 days. As far as I could find, the UK and EU have no equivalent single switch. You file per company.

The abuse chain

The FTC says job and employment agency scam reports tripled from 2020 to 2024, with reported losses rising from $90 million to $501 million. In the UK, as relayed from the Guardian in April 2026, Report Fraud logged more than double the recruitment-scam reports in 2024 compared with 2022, Lloyds saw a 237% rise from January to August 2025. The same coverage says scammers clone recruiter profiles and tailor messages using CV details, and AI removes the bad grammar that used to give them away.

I found no source linking a specific scam campaign to a specific broker breach. The link is structural: the same fields, the same buyers.

SOURCE Public professional profile. Name and job title.

↓

SCRAPER Fake accounts and bulk pulls, as alleged in the ProAPIs case.

↓

GATE 1 / You Unique alias and separate number per signup. Article 15 and 17 requests. DROP if Californian.

↓

ENRICHMENT VENDOR Adds email and mobile number, the KASPR model.

↓

GATE 2 / Company Vendor due diligence. Ask every recruiter where they got the data.

↓

BROKER AND BREACH Resold, then leaked. NPD, DemandScience, the open 4.3 billion database.

↓

ABUSE Spam and cold calls. Smishing and vishing. Fake recruiter offers.

↓

GATE 3 / Both Call back on a published number. Phishing-resistant MFA. Port-out PIN. Forward texts to 7726.

The same list hits staff. Google's Threat Intelligence Group described a group phoning employees as fake IT support in 2025, walking them into approving a rogue app that pulled Salesforce data. Google's own instance held business contact details. Cifas reported unauthorised SIM swaps up 1,055% in 2024, nearly 3,000 cases, and Which? found weak identity checks at some networks in April 2026.

What to do

Individuals:

Use a fresh email alias and, where you can, a separate number for each signup. When spam arrives, the alias names the leaker.
Register with TPS (CTPS for companies and some partnerships). It covers live marketing calls and criminals ignore it.
Forward scam texts to 7726 and report fraud at Report Fraud, which replaced Action Fraud on 4 December 2025.
Send an Article 15 request asking where a firm got your number, then Article 17 erasure. The ICO says firms have one calendar month. Complain to the ICO if they stall.
Ask your network for an extra PIN on account changes, and prefer passkeys to SMS codes.
Never trust an inbound recruiter call. Hang up and ring the company's published number. Anyone asking for ID or payment is hostile.

Companies:

Staff contact data is part of the attack surface. Require out-of-band verification for helpdesk resets and deploy phishing-resistant MFA. Brief people on recruiter-themed lures. Audit which agencies and enrichment tools receive employee data. Under UK GDPR, an agency sourcing candidates is usually its own controller, but read the contract. Ask for their Article 14 notice and retention period, then ask where the data came from.

What I could not confirm

No evidence of resume watermarking for leak tracing. No named scam wave tied to a named breach. The Eightfold claims are unproven. Have I Been Pwned marks the NPD data unverified, and its size varies by outlet. DemandScience phone numbers come from one outlet. The Guardian figures reached me second-hand. I found no ICO enforcement over recruitment scraping, only the audit.

Verdict

The recruiting industry treats your contact details as inventory, and inventory leaks. Regulators have started charging for it, but a fine in France does not stop a call to your mobile.

So defend like the number is already out. Make the number traceable, and make every inbound recruiter earn trust through a channel you chose.

Sources

Elusive Thoughts

30/10/2026

A recruiter, a hacker and data broker you never spoke