Who is liable if our AI screening tool rejects a real candidate or approves a fake one?

You are. The deploying firm owns the decision, not the software vendor. Read your vendor contract and you will find language that pushes liability back to you. That means if your tool screens out a protected class or waves through a fraudulent hire who then damages your client, the claim lands on your desk. Keep a human in the loop on every reject and every final submit, and document why.

How do we spot an AI-generated candidate without slowing down every submit?

You do not check everyone the same way. Set a risk tier. Full remote roles, tech contracts, and high-pay reqs get the full verification stack. Local, in-person, lower-risk roles get the light version. Most fraud clusters around remote and high-value roles, so put your effort there. A live video call with two unscripted questions catches most fakes in under five minutes.

Is this really a bigger problem than the talent shortage?

According to GoodTime's 2026 report, yes, for the first time. Fake candidates now rank as the number one hiring challenge. It does not mean the talent shortage went away. It means fraud got easy and cheap, and volume exploded. One bad placement that turns out to be a fake or a proxy can cost you a client relationship you spent years building. That math makes prevention worth the effort.

AI Fraud Firewall: Stop Fake Candidates in Staffing

AI-generated fake candidates just passed talent shortage as the top hiring challenge. Gartner says one in four profiles will be fake by 2028. Here is a concrete protocol to build your fraud firewall now, before a bad hire lands on your client's desk.

AI-generated fake candidates are now the number one hiring challenge, according to GoodTime's 2026 Hiring Insights Report. For the first time, they beat out the talent shortage. Gartner projects that one in four candidate profiles will be fake by 2028.

This is not a future problem. Recruiters are already interviewing people who do not exist, running background checks on stolen identities, and submitting proxy candidates who let someone else do the actual work after placement. The tools to fake a resume, a headshot, a voice, and even a live video interview are cheap and getting cheaper.

Here is the part most VPs miss. When your AI screening tool makes the wrong call, you own it. Not the vendor. The liability for an AI screening decision rests with the firm that deployed the software. So a vendor promise is not a defense. You need an operational protocol you control.

Why Your Client's Desk Is the Wrong Place to Catch This

By the time a fake candidate reaches your client, the damage is already priced in. Your client interviewed a ghost, wasted their hiring manager's time, and now questions every profile you send after that. One fake candidate does not cost you one placement. It costs you the client's trust in your entire pipeline.

The fraud firewall belongs at your front door, not theirs. Your job is to catch the fake before it ever hits their inbox. That means building verification into your intake and screening steps, not bolting it on after a client complains.

Most firms are not doing this. They are relying on gut feel, a quick phone screen, and a background check that only confirms the identity on paper, not the person on the call. That gap is exactly where fraud lives.

The Four-Layer Fraud Firewall

You do not need a six-figure tool to start. You need a repeatable protocol your recruiters run every time. Build it in four layers.

Layer 1: Identity at Intake

Verify the human is the human before you spend an hour on them. Require a live video call for any remote or high-value role. Ask two questions that are not on the resume and cannot be pre-scripted. Watch for lag between your question and their answer, audio that does not match lip movement, and eyes that keep drifting to a second screen.

Real candidates answer messy, specific questions fast. Fakes and proxies stall, deflect, or read.

Layer 2: Resume and Profile Cross-Check

AI writes clean resumes. Too clean. Look for roles with no verifiable company footprint, employment dates that line up perfectly with no gaps ever, and skill lists that read like a job posting instead of a career. Run the candidate's name and photo through a reverse image search. A stock or AI-generated headshot is a fast tell.

Cross-check the LinkedIn profile against the resume. Fabricated profiles often have thin connection counts, no engagement history, and a creation date that is suspiciously recent.

Layer 3: Skills Verification You Watch Live

Take-home tests are dead for fraud detection. Anyone can paste your test into a chatbot. Instead, run a live skills exercise where you watch the work happen. Screen share. Ask them to talk through their thinking. For a developer, have them debug live. For an accountant, walk a real scenario.

Proxies fall apart here because the person on the call is not the person who can do the work.

Layer 4: Human Sign-Off on Every Submit

No profile goes to a client without a named recruiter signing off that they ran the firewall. Keep a short record for each submit: video verified, profile cross-checked, skills confirmed, date, recruiter name. This does two things. It catches fraud, and it protects you legally when a screening decision gets questioned.

Do not let the AI make the final call alone. A human decides. The AI flags and speeds up the work. That distinction is your legal firewall as much as your fraud firewall.

Set Risk Tiers So You Do Not Slow Everything Down

You cannot run four full layers on every submit or your recruiters will revolt. So tier your risk.

High risk: Fully remote roles, tech and engineering contracts, any role paying above a threshold you set, and anything with client access to sensitive data. Full four-layer firewall.
Medium risk: Hybrid roles and mid-pay positions. Video verification plus profile cross-check.
Low risk: Local, in-person, lower-pay roles. Video verification only.

Fraud clusters in remote and high-value roles because that is where the payoff is. Put your effort where the risk lives. This keeps your speed on the roles that do not need heavy screening.

Where AI Helps and Where It Hurts You

AI is useful here. It can flag reverse-image matches, scan for profile inconsistencies, and detect video artifacts faster than a person. Use it as a first filter that raises flags.

What AI cannot do is take the final decision off your plate. The moment you let a tool auto-reject or auto-approve, you own every mistake it makes and you have no human record to defend the call. Vendors will tell you their model is accurate. Maybe it is. But accuracy is not the same as liability, and the contract already tells you who holds the liability.

Use AI to do the fast, boring detection work. Keep the judgment with your recruiters.

Do This Week

Pull your last 20 submits for fully remote roles. For each one, ask a simple question: did a live human on our team verify this candidate on video, and do we have a record of it? If the answer is no on more than a handful, you have a hole in your front door.

Write down the four layers above as a one-page checklist. Assign one person to own it. Start running it on every high-risk submit this week. You do not need a vendor to start. You need a protocol and a name attached to it.

The firms that build this now protect their client relationships before the 2028 numbers arrive. The ones that wait will explain to a client why they sent a candidate who does not exist.

The AI Fraud Firewall: Stopping Fake Candidates Before They Reach Your Client