ServiceNow's June 11 Cut Just Untethered Bengio's Montreal Bench
ServiceNow's June 11, 2026 layoff and AI-efficiency freeze put ex-Element AI researchers in play. Source them by arXiv, not LinkedIn title.
On June 11, 2026, ServiceNow confirmed a three-figure layoff and framed it as "real AI efficiencies." The named cuts hit solution consulting, sales, product marketing, and L&D. Research was not named, which is exactly why the Blind chatter from ServiceNow Research staff (the team that came in through the 2021 Element AI acquisition) is the signal worth acting on this week.
If you run ML hiring, this is the rare moment when a hard-to-reach research bench becomes warm to outreach. The catch: you will not find them with a LinkedIn title search. Their employer field says "ServiceNow." Their team name, "ServiceNow Research" or the Montreal AI Innovation Hub, almost never appears in their headline. "Element AI" returns near-zero. The roster lives on arXiv.
What actually happened on June 11
ServiceNow told NowBen the restructure is meant to "grow sustainably and win" and that the company is "actively investing in and hiring for the AI-focused skills this era demands, while managing headcount with discipline to end the year where we started." This is the first cut since CEO Bill McDermott's 2023 "no job cuts" pledge, which is what makes the reversal newsworthy beyond the headcount.
The June 11 event is the third step in a sequence, not a one-off:
- April 22: McDermott confirmed ServiceNow will not backfill natural attrition through year-end, banking on AI productivity to hold 2027 headcount flat.
- April: The QE (quality engineering) function was eliminated entirely, not shrunk. That precedent matters.
- June 11: The first explicit layoff, attributed to AI efficiencies.
The QE elimination is the leading indicator. When a company is willing to remove an entire function rather than trim it, every non-revenue function reads the memo. ServiceNow Research is a non-revenue function. The Blind posts from that team are not paranoia; they are pattern recognition.
Why this specific pool matters
ServiceNow closed its acquisition of Element AI on January 8, 2021 for approximately US$230 million. The deal was unusual: ServiceNow kept the research scientists and the patents and effectively abandoned the business. Roughly half of Element AI's workforce, including the "vast majority" of corporate employees, was terminated shortly after close. ServiceNow agreed to pay an additional US$10 million to retain key employees and consultants, including Yoshua Bengio (Turing Award 2018, technical advisor) and co-founder Nicolas Chapados.
That US$10 million retention pool was paid at close. Five years on, mid-2026, any acquisition-era RSU cliffs have fully vested. The financial handcuffs that kept this group in place through 2024 are gone. Add a public AI-efficiency layoff and a non-revenue research mandate, and the math on staying versus listening to a recruiter changes in one quarter.
The retained team became ServiceNow Research, anchored in Montreal as the AI Innovation Hub. Its recent output is a tight thematic cluster: web agents ("How to Train Your LLM Web Agent: A Statistical Diagnosis," NeurIPS 2025), XC-Cache for LLM inference, and BrowserGym / AgentLab-style evaluations. If your team is building agents, evals, or RL infrastructure, this is the bench.
The title is the wrong primitive
Here is the operational problem. Our index of professional profiles surfaces only a tiny handful of people in the Montreal / Quebec region whose current title explicitly contains "ServiceNow Research." A boolean string built on that phrase, or on "Element AI," will miss almost the entire pool. The team appears on LinkedIn as:
- "Research Scientist, ServiceNow"
- "Applied Research Scientist, ServiceNow"
- "Senior Research Engineer, ServiceNow"
- "Staff Research Scientist, ServiceNow"
None of those strings differentiate a ServiceNow Research author from a platform engineer in Santa Clara. The arXiv author list for any 2023 to 2025 ServiceNow Research paper is a more complete and more current roster than LinkedIn.
The arXiv author list for any 2024 ServiceNow Research paper is a more complete roster than LinkedIn will ever give you.
This is the broader lesson for sourcing ML research engineers in 2026: at companies where research is a cost center inside a platform business, the public record (papers, GitHub, workshop talks) is denser than the HR record. Title-based boolean assumes the company is the unit of identity. For this pool, the paper is.
That inversion is the reason we built Refolk. You describe the person in plain English ("ML research scientists in Montreal who have co-authored web-agent or BrowserGym papers in the last 18 months, dual-affiliated with Mila"), and you get a ranked shortlist pulled from GitHub, LinkedIn, and the open web together, not LinkedIn alone. The point is not magic. The point is that the right primitive for this pool is "published with," not "titled as."
A paper-trail sourcing recipe
Here is the concrete workflow. None of it requires a paid database.
1. Start at the publication page
ServiceNow Research lists its papers at servicenow.com/research/publication.html. Pull every paper from 2023 onward. Extract every distinct author with a ServiceNow affiliation. You will hit roughly fifty to eighty names, depending on how you handle interns and visiting researchers. Recurring senior authors to anchor on:
- Nicolas Chapados, Element AI co-founder and chief science officer
- Dzmitry Bahdanau, co-author of the original attention mechanism paper
- Alexandre Lacoste, Alexandre Drouin, David Vazquez, Issam Laradji, recurring senior authors on the web-agent and LLM-tooling work
- Pau Rodriguez, Sai Rajeswar Mudumba, Perouz Taslakian, Maxime Gasse, Massimo Caccia, Krishnamurthy DJ Dvijotham, Christopher Pal
2. Cross-reference Mila
Most senior ServiceNow Research scientists are dual-affiliated with Mila - Quebec AI Institute, HEC Montreal, University of Waterloo, or UBC. Mila publishes its member directory. The intersection of "ServiceNow Research author" and "Mila member" is your highest-signal subset, because those names update their Mila page faster than their LinkedIn.
3. Use co-authorship as a graph
Once you have ten anchors, pull every co-author on their last three papers. Filter to Montreal-resident, ServiceNow-affiliated. You will discover senior ICs who never show up in a title search because their headline is something like "PhD, ML." This is the part of the workflow that most recruiters skip and where the actual hidden bench lives.
4. Reach them on their channel
Mila seminars, Bengio's lab social graph, and the Montreal AI Symposium (MAIS) author lists are channels recruiters in SF and NYC rarely use. A short, specific message that references a 2024 paper outperforms a generic "I saw your profile" message by an order of magnitude with this group. They are academics first.
Why the timing is unusually good
Three forces stack this quarter:
The macro story gives you cover. According to outplacement firm Challenger, Gray & Christmas, AI was the most-cited reason for layoffs across every industry for the third month running. That means a passive candidate hearing from you in July 2026 is not surprised by the outreach; they expected it. Reply rates on warm DMs are higher than the same message would have produced in Q4 2025.
ServiceNow's cut is a public AI-displacement event. TechJack Solutions called it "one of the clearest direct AI displacement events to reach the public record from a major enterprise software company." That framing is now in the candidate's information set. They know what game they are in.
The research function is structurally exposed. When a CEO attributes cuts to "real AI efficiencies," non-revenue research is the politically obvious place to look next, even when it is not in the current round. The Blind posts reflect that read. Warm outreach now lands on people who are quietly already at the "what would I do next" stage but have not started a search. They will not be in the open market in ninety days; they will already be talking to someone.
What to build around them
If you are an engineering leader, the practical question is what team this pool slots into. The thematic cluster (web agents, BrowserGym, AgentLab, LLM inference caching, RL evals) maps almost one-to-one to the agent-platform roadmaps at Sierra, Adept, Imbue, Cohere, and the agent teams at Anthropic and OpenAI. The Bengio Element AI alumni network is also overrepresented at Cohere (Toronto) and at Recursion's ML org, so do not assume your competition is only US-based.
For startups that cannot match Big Lab compensation: the lever is publication freedom plus compute, not base salary. This group left ServiceNow Research mentally before they leave it operationally because the research mandate inside an ITSM platform never fully made sense to them. A clear externalizable roadmap matters more than another $30K of base.
If you are running this search now, the cheapest accelerant is to stop spending hours hand-mapping author lists. That is the part Refolk collapses. You ask in plain English for ServiceNow Research authors with a specific paper cluster and Mila affiliation, and the cross-source matching against GitHub, LinkedIn, and arXiv happens in one query instead of three afternoons.
The wider pattern
ServiceNow is not unique. Any platform company that bought a research lab in the 2019 to 2021 window is now five years past the acquisition retention package and one AI-efficiency narrative away from putting that lab in play. The sourcing playbook is the same: ignore the title, pull the papers, walk the co-authorship graph, message through the academic community.
For ServiceNow Research talent specifically, the window is this quarter. After the next earnings call, this pool will either be reabsorbed into a renamed "AI Platform" group (which buys them another year of inertia) or split across three other labs that move faster than your pipeline. June 11 was the signal. Act on it before September.
FAQ
How many people are actually in the ServiceNow Research / ex-Element AI pool?
ServiceNow has not disclosed the figure. Based on the author lists of papers published between 2023 and 2025 on the ServiceNow Research publication page, the active research roster looks like fifty to eighty distinct authors with primary ServiceNow affiliation, weighted heavily to Montreal and concentrated around the original Element AI technical staff that was retained at the January 2021 close.
Are the June 11 cuts going to hit research next?
ServiceNow has not said so. The June 11 cuts publicly hit solution consulting, sales, product marketing, and L&D. The reason the Blind posts from ServiceNow Research staff are meaningful is not that research has been told it is next, but that the April QE elimination established a precedent for cutting whole non-revenue functions, and the company is attributing cuts to AI efficiencies. That combination is what makes warm outreach land.
Why not just search "Element AI" on LinkedIn?
Because almost nobody on this team lists Element AI as their current employer. The acquisition closed in January 2021. Their current employer field says "ServiceNow." Their team name rarely appears in their headline. A title or company string returns a tiny fraction of the real bench. The arXiv author list for any recent ServiceNow Research paper is a denser source than LinkedIn for this specific pool, which is the core argument of this piece.
What about Bengio himself?
Yoshua Bengio joined as ServiceNow technical advisor at the close of the Element AI acquisition and remains the gravitational center of Mila and Montreal AI broadly. He is not the hire. He is the social-graph anchor: most senior ServiceNow Research scientists trace back to his lab or to Mila co-authorships, which is why Mila-based channels (seminars, MAIS, lab alumni networks) outperform cold LinkedIn for reaching this group.