Meta's 1,500 MCI Signers Are a Pre-Built Shortlist. The May 20 Layoffs Made It Live.
More than 1,500 Meta engineers signed the anti-surveillance petition. Then 8,000 got laid off and 7,000 reassigned into the AI org they protested. Source them off LinkedIn.
Most sourcing lists are inferred. This one is signed. More than 1,500 Meta employees put their names on a petition opposing the Model Capability Initiative weeks before the company laid off 8,000 people and reassigned 7,000 more into the exact AI org the petition was protesting. If you recruit privacy-conscious engineers, this is the cleanest values-aligned cohort the industry will produce in 2026.
What MCI actually is, and why the signers matter
Reuters first reported the Model Capability Initiative on April 21, 2026. The tool, also known internally as the Agent Transformation Accelerator (ATA), was deployed on April 22 to US Meta employees' work laptops. It captures mouse movements, keystrokes, clicks, and periodic screenshots across hundreds of approved work applications and websites. The stated purpose: generate training data for Meta's agentic AI.
US employees got no opt-out. European Meta employees were entirely exempt because GDPR and national worker-protection laws in EU member states require explicit employee consent. That asymmetry, more than the surveillance itself, is what radicalized the workforce. American engineers watched their European colleagues opt out by default while their own keystrokes shipped to a training corpus.
Opposition crystallized into a formal petition. It crossed 1,000 signatures fast and topped 1,500, per the BBC. The petition cited the US National Labor Relations Act directly: "Workers are legally protected when they choose to organize for the improvement of working conditions." That language matters for search. Anyone quoting it on a personal blog, a Mastodon post, or a Bluesky thread is, by definition, a signer or an adjacent organizer.
The May 20 layoffs hit the same week
On May 20, 2026, Meta began notifying roughly 8,000 employees of layoffs. Singapore went first at 4 a.m. local time, then the UK and US as their mornings began. Chief People Officer Janelle Gale simultaneously announced that upward of 7,000 workers would be redirected into newly created AI-focused teams: Applied AI Engineering, Agent Transformation Accelerator XFN, and Central Analytics. Meta also cancelled 6,000 open requisitions, bringing the effective headcount reduction to 14,000 positions.
Read those org names again. The reassignment org is named after the surveillance program. A petition signer who got moved into Agent Transformation Accelerator XFN is now building the agents MCI was meant to train. That is not a coincidence; that is a values trap. Those engineers still have a paycheck, which means they will not post #OpenToWork, which means LinkedIn will not surface them. They are the highest-intent passive candidates of the year and the hardest to find with conventional tooling.
Where the WARN filings put the laid-off cohort
The signers who did get cut are geographically concrete. California WARN Act filings confirm 124 positions at Meta's Burlingame office effective May 22 and 74 at Sunnyvale effective May 29. A Washington State filing surfaced via Blind covers roughly 1,395 layoffs across four WA facilities including Dexter Ave N Seattle and Bellevue, effective July 22, 2026. At least 318 of the layoffs hit Menlo Park headquarters per a separate California WARN notice. Meta's Superintelligence Labs workforce now sits at just under 3,000 employees.
Severance gives you the calendar. US workers in the May round received 16 weeks of base pay plus two additional weeks per year of service, with 18 months of health coverage. That puts most affected ICs on severance through roughly September 2026. Your active-search window opens now and closes when the runway does.
Why LinkedIn is the wrong starting point
Two reasons. First, the reassigned 7,000 are still employed, so their profiles say "Meta" and their headlines say nothing useful. Second, the laid-off cohort skews ML and DS, where the most senior engineers maintain deliberately sparse profiles. The Blind thread on the May layoffs put it bluntly: "Every single LinkedIn post I have seen so far of people who were laid off from Meta are from people who were either Machine Learning Engineers or Data Scientists." Those are exactly the people the signers were.
The higher-signal surfaces are public but fragmented:
- The petition itself and its adjacent flyers. Employees taped flyers in Meta's California and New York offices reading "Don't want to work at the Employee Data Extraction Factory?" That phrase surfaces signers on Twitter, Bluesky, and Mastodon. Search it verbatim.
- GitHub commit graphs. PyTorch, Llama tooling, FAIR repos, and the long tail of personal repos by Meta engineers. Recent activity changes plus a values-loaded README is a tell.
- UTAW membership. United Tech and Allied Workers, a division of the UK Communication Workers Union, is the active organizing home for ex- and current Meta UK staff running their own unionization push.
- More Perfect Union's audience graph. The labor advocacy group published the leaked April 30 all-hands audio and functions as a de facto media organ for the petition cohort.
- Blind threads. Verified Meta emails are required to post. Layoff threads contain self-disclosed team, level, total compensation, and severance terms. That is qualified outbound, not gossip.
This is exactly the kind of cross-surface query that breaks Boolean. You are looking for "someone who committed to a public ML repo in the last 90 days, has a Meta tenure of at least three years, and has posted about worker organizing or privacy on a non-LinkedIn surface." That is a plain-English query, not a search string, which is why we built Refolk: you describe the person in English and get a ranked shortlist across GitHub, LinkedIn, and the open web, including the surfaces above.
The four sourceable tiers
Not every signer is equally reachable. Sort them.
Tier 1: The laid-off ML and DS signers on severance
Roughly 8,000 laid off, with disproportionate ML and DS exposure. Severance runs through about September 2026. These are the people writing post-mortem Substacks, doing one-month sabbaticals, and taking exactly one round of conversations before signing. Get to them in June and July or pay a premium in August.
Tier 2: The reassigned 7,000 stuck in ATA
Same paycheck, worse job, signed petition opposing the org they now sit in. They will not post #OpenToWork. They will reply to a specific, named outreach that demonstrates you read the petition. Refolk surfaces this group because it cross-references current Meta tenure with off-LinkedIn signal, which is the only way you find someone whose LinkedIn still says Applied AI Engineering.
Tier 3: The UK and EU cohort
European Meta engineers were exempt from MCI by GDPR. They watched the US cohort get surveilled, then laid off. UTAW organizers in the UK are the most legible, values-aligned EU pool for any US or UK startup that wants to plant a privacy-first flag. Membership rolls are not public, but UTAW's branch meetings, panel appearances, and public statements are.
Tier 4: The FAIR diaspora
The values mismatch was structural, not just rhetorical. Yann LeCun, Meta's chief AI scientist for 12 years, departed in November 2025 after being asked to report to Alexandr Wang. "You don't tell a researcher what to do," he told the Financial Times. "You certainly don't tell a researcher like me what to do." He raised $1 billion to found AMI Labs in Paris, drawing the founding team "almost entirely from Meta's AI research organisation." Soumith Chintala, PyTorch's creator, is now CTO at Mira Murati's Thinking Machines Lab. The pattern is set. Senior FAIR people land at well-funded labs run by people they already respect.
The petition is a public artifact, not a private grievance. Signers self-selected on the record. </pull>
If you are recruiting for a smaller lab, you are not competing with AMI or Thinking Machines for the top 20 names. You are competing for the 1,480 signers nobody is bidding on yet.
How to actually run the outreach
Three things to get right.
Lead with the petition language, not the layoff
"I saw you signed the MCI petition" is a non-starter. You do not know who signed. What you do know is the public posture: NLRA-cited organizing, opposition to non-consensual training data extraction, EU exemption envy. Lead with the specific product or principle that maps. "We do not log keystrokes, ever, and our model training data is opt-in by default" is a sentence that, to this cohort, reads as a recruiting pitch.
Skip the severance math in the first message
It is tempting to anchor on "your severance ends in September." Do not. The signers know. Mentioning it reads as predatory. Anchor instead on the technical work and the org's stance on data and consent. The clock takes care of itself.
Treat reassigned signers as a separate funnel
Reassigned signers need a longer ramp. They are not job-hunting yet because the paycheck is intact, but they will be by Q4. Build a relationship in June, send a specific role in September. This is where sourcing Meta engineers off-platform actually pays off, because the people you want are not yet doing the things that flag intent on LinkedIn.
The broader pattern this fits
Meta is not the only large AI org training agents on employee behavior. It is the only one where 1,500 employees told you, in writing, that they will not ship the product. For sourcing privacy engineers and ex-Meta AI engineers, that signal is rarer and more durable than any layoff list. It will still be useful in 2027, because the people who signed in 2026 are the same people who will sign the next one at the next company.
The work, then, is not to scrape the petition. The petition is not scrapeable, and treating it as a leak misreads the room. The work is to recognize the petition as one of several public artifacts (flyers, GitHub repos, UTAW posts, More Perfect Union coverage, Blind threads, the Stephane Kasriel walk-back memo) that, taken together, form a graph. Refolk's job is to walk that graph for you in plain English, so you spend your time on outreach instead of Boolean.
The Meta May 2026 layoffs cleared 14,000 positions when you include the cancelled reqs. Of those, the 1,500 signers are the cohort you can actually identify, qualify, and reach before September. The other 12,500 will route through normal channels. The signers will not.
FAQ
How do I find petition signers if the petition itself is not public?
You triangulate. Search the flyer language ("Employee Data Extraction Factory") and the NLRA quote on Bluesky, Mastodon, and personal blogs. Cross-reference UTAW UK public statements, More Perfect Union coverage credits, and recent GitHub activity from accounts tied to Meta email domains. Refolk does this triangulation in one query, returning a ranked list with the evidence trail attached.
Are reassigned engineers really worth recruiting if they have not quit?
Yes, and they are arguably the highest-value tier. They signed a petition opposing the org they now sit in, which is the strongest possible signal of misalignment without the cost of an active job search. They will not post #OpenToWork. A specific, well-framed outreach in June lands a September conversation. The reassigned 7,000 are the cohort no layoff tracker catches.
What about the European Meta engineers who were exempt from MCI?
They are the parallel cohort. GDPR kept them out of MCI, but UK organizers are running their own UTAW unionization campaign and watched the US layoffs hit colleagues who shared their values. For any US or UK startup hiring in London, Dublin, or Paris, this is the cleanest privacy-first EU sourcing pool available right now.
How long is the window?
Severance for the May 20 round runs roughly through September 2026 (16 weeks base plus tenure). The Washington State WARN filings show separations effective July 22. The active-search window for Tier 1 is June through August. Tier 2 (reassigned) opens in Q4. Tier 4 (senior FAIR) is largely spoken for by AMI Labs and Thinking Machines, so do not lead with it.