xAI Pre-Tested 500 Grok Tutors Before Firing Them. Source the Scores.

On the evening of September 12, 2025, xAI cut roughly 500 people from the team that trains Grok, about a third of a 1,500-person annotation org. The unusual part is not the number. It is that xAI made every one of them sit an internal skills test in coding, finance, and social media behavior first, then used the results to decide who stayed. If you recruit for an AI lab or an RLHF vendor, that layoff list is the closest thing to a pre-scored shortlist you will ever get for free.

What actually happened at xAI on September 12

xAI laid off about 500 generalist "AI tutors" (its internal term for the humans who train Grok) on the night of September 12, 2025, after running them through mandatory skills assessments in coding, finance, and social media behavior. The main annotator Slack room dropped from over 1,500 members that Friday afternoon to just over 1,000 by evening, giving a real-time headcount signal that reporters could screenshot.

The severance economics matter for your outreach timing:

• Access to company systems was killed the day of the notice.
• Pay runs through the end of contract or November 30, 2025, whichever comes first.
• xAI simultaneously announced a "10x" expansion of its specialist tutor team in STEM, finance, medicine, and safety.

Read that again. Same company, same week, fired 500 generalists and pledged to grow a tenfold-larger specialist bench. The tests were the sorting mechanism between the two groups. That is not a distress signal for the fired cohort. That is a filter someone else paid to run.

500

Why this cohort is a pre-vetted shortlist, not a distress list

These 500 people are the only tranche of AI trainers on the open market who just took a rigorous, employer-run domain assessment in the last 90 days. Treat them as involuntarily scored talent, not as generic layoff casualties. The mechanism is simple: xAI paid for the assessment, the market gets the output.

Two nuances that most recruiters will miss:

1. "Generalist" at xAI is not generalist elsewhere. xAI's own live "AI Tutor, Finance Specialist" job description asks tutors to teach models how people approach discussions in finance by gathering or providing text, voice, and video data, sometimes with annotations. That is closer to what Scale AI calls a "domain contributor" than to a labeler. Many of the 500 are subject-matter experts who scored below a raised bar, not unskilled clickers.
2. The survivors are the elite, but the cut cohort is not the bottom of the labor market. Before the layoff, employees were called into one-on-one meetings to walk through duties and flag colleagues they wanted to recognize. The 500 came out worst against xAI's own rubric, in a team that xAI had already been hiring "experts from fields including finance, science, accounting, and comedy" into. That is a much higher floor than a Mechanical Turk pool.

If you are staffing a specialist tutor bench at Anthropic, Meta Superintelligence, Mercor, Invisible, Surge AI, or Scale AI, this is the rare moment where the filter has already been run for you and the candidates have a hard cash deadline.

The one reason these 500 are individually findable

Because xAI built annotation in-house instead of outsourcing to Scale, Surge, or Outlier, these workers list "xAI" directly on LinkedIn instead of a vendor name. That is the unlock. Most annotation labor is anonymous inside vendor rosters; this cohort is not.

xAI's in-house approach, as reported, "gave the company tighter control over quality and a direct relationship with its trainers, but meant xAI shouldered the full recruiting and onboarding burden itself." For sourcing purposes, that translates to: 500 W-2 or direct-contract workers whose employer field on LinkedIn, GitHub bios, and personal sites reads xAI, not a staffing agency.

This is the exact gap Refolk closes. You describe the person in plain English ("US-based AI tutor or data annotator who listed xAI as employer in 2024 or 2025, with domain background in finance, medicine, coding, or safety") and get a ranked shortlist across LinkedIn, GitHub, and the open web, without stitching Boolean strings across three tabs.

The size of the pool you are fishing in

In Refolk's index of professional profiles, there are about 5,056 US-based people whose current or recent title matches AI Trainer, Data Annotator, AI Tutor, or Data Labeler. The xAI cohort is roughly 9.9% of the entire searchable US pool, concentrated in a single employer, on a single date.

Two things fall out of this table. First, if you land even 30 of the 500, you have moved almost 0.6% of the entire US AI-tutor labor market into your org in one quarter. Second, the top current employers of active US annotators in Refolk's index (DataAnnotation, Outlier, Alignerr, Handshake) are your competition. They will re-absorb xAI alumni faster than you will if you wait past Halloween.

9.9%

The Boolean xAI already wrote for you

Use xAI's own test domains as your filter, because they map one-to-one onto what the survivors were kept for and what the "10x" specialist expansion is targeting. The four buckets are coding, finance, social media behavior and internet culture, and safety, with STEM and medicine added on the specialist side.

A practical query, in plain English, looks like this:

• Present or past employer contains "xAI" or "x.ai" between 2024 and 2025.
• Title contains AI Tutor, AI Trainer, Data Annotator, RLHF, Prompt Engineer, or Model Trainer.
• Background signal in one of: quantitative finance, licensed clinician, published researcher, senior software engineer, trust and safety, or professional comedy and writing.
• US-based, available in the next 60 days.

The "comedy and writing" bucket is the one nobody else will search. xAI had been hiring writers to shape Grok's personality, and those profiles will not show up under any standard RLHF search string. They will show up under stand-up credits, Substack bylines, and staff-writer roles at defunct digital media brands.

Where the xAI alumni will actually land

Expect the 500 to re-cluster inside four vendors within 30 days, based on who currently employs the largest concentrations of US AI tutors. Watch these employer pages, not job boards.

• DataAnnotation and Outlier will absorb the generalist tail fastest, because their onboarding is self-serve and their pay-per-task model matches the cash-cliff urgency.
• Alignerr and Handshake (which runs a large university-adjacent tutor program) will pick up the more credentialed slice.
• Mercor and Invisible will target the domain specialists, because their pitch to enterprise buyers is "vetted expert supply," which is exactly what xAI's internal test just certified.
• Scale AI and Surge AI will circle the safety and coding subsets, because those tie into their existing frontier-lab contracts.

The 500 are not a distress list. They are a filter xAI paid to run, and the survivors are not on the market.

If you are hiring against these vendors, the sourcing move is to set a saved query for profile updates where the employer field changes from xAI to any of the eight names above, over the next eight weeks. That is your poach list for 2026.

Who signs the offer on the other side

Diego Pasini now oversees the annotation unit at xAI, which makes him the person to watch for the "10x specialist" build-out and, indirectly, the internal reference for what "good" looked like on the tests. On the departing side, xAI's finance chief Mike Liberatore left around the end of July after only a few months, which is useful context if a candidate cites organizational chaos as the reason they were open to interviewing.

By mid-2026, xAI paused specialist tutor hiring entirely, not because budget dried up, but because recruiting operations became strained processing incoming candidates. Read that carefully. Demand for these skills did not fall. Throughput did. If you can process a pre-filtered shortlist faster than xAI can process a cold funnel, you win the cohort. That is the specific friction where a plain-English sourcing tool like Refolk pays for itself, because the constraint is not "find the people," it is "find them before Nov 30 and get to outreach in a week, not a quarter."

A 14-day playbook against the Nov 30 cliff

Run these steps in parallel starting the Monday after you read this. The window is short and the competition list is public.

1. Day 1 to 2. Pull the ex-xAI tutor list. Segment into four buckets: coding, finance, safety, and STEM/medicine. Flag anyone with a domain credential (CFA, MD, PhD, senior IC title before xAI).
2. Day 3 to 5. Cross-reference against GitHub for the coding subset. The ones with recent commits on eval harnesses, prompt libraries, or model-grading repos are your top decile.
3. Day 6 to 8. First outreach. Lead with the domain, not the layoff. "I'm hiring finance specialists to train models on earnings calls and 10-Ks" beats "saw you were impacted."
4. Day 9 to 11. Structured screen. Do not re-run xAI's test. Ask them to describe the rubric they were graded on and what they think the survivors did differently. The answer is a better signal than any take-home.
5. Day 12 to 14. Offer. Match or beat their xAI hourly. The candidates know their old rate and they know the cliff date. Offer letters that land before November 15 close; letters that land after Thanksgiving do not.

The teams that will lose this cohort are the ones still writing Boolean strings on November 25. The teams that will win it decided who they wanted by October 10 and spent the rest of the month on conversations, not searches.

FAQ

How do I identify the specific 500 xAI tutors on LinkedIn?

Filter for current or past employer "xAI" or "x.ai" with a title containing AI Tutor, AI Trainer, Data Annotator, or Model Trainer, then narrow by "started 2023-2025" and US location. Because xAI ran annotation in-house rather than through Scale or Surge, these workers list xAI directly, which is unusual for the annotation labor pool and is the reason this cohort is individually findable at all.

Are the laid-off xAI tutors actually skilled, or is this a low-end labor pool?

They are more skilled than the "annotator" label suggests. xAI's own job descriptions ask tutors to generate domain content in finance, medicine, and coding across text, voice, and video, and the company had been hiring from finance, science, accounting, and comedy backgrounds. The 500 scored below a raised bar on an internal test, but the bar itself was set for specialists, not clickers.

Why does the November 30 date matter so much?

Because xAI killed system access on September 12 but pay runs through November 30, candidates have both severance runway and a hard deadline in the same 11-week window. That combination is rare in AI-labor layoffs (Scale AI's 2024 contractor purges had neither) and it means offers landing before mid-November convert at far higher rates than offers landing in December.

Which employers will re-absorb this cohort the fastest?

Refolk's index shows the largest concentrations of active US AI tutors sit at DataAnnotation, Outlier, Alignerr, and Handshake, so those are the fastest re-absorption paths for the generalist tail. Mercor and Invisible will move on the domain specialists because their enterprise pitch depends on vetted expert supply, and Scale AI and Surge AI will target the safety and coding subsets tied to their frontier-lab contracts.