The Real Ex-TikTok Trust & Safety Engineer Pool Is Under 50

Every T&S recruiter at Discord, Roblox, Character.AI, and the DSA-regulated platforms just opened LinkedIn and typed "ex-TikTok trust and safety." That search returns roughly 40,000 people, and almost none of them build rule engines.

In early July 2026 TikTok confirmed roughly 300 Dublin roles at risk plus cuts in Singapore and Indonesia, layered on top of the earlier dismantling of its ~150-person Berlin moderation team. The layoff coverage is loud. The actual addressable engineer pool is small enough to fit on one screen.

The real engineer pool is under 50, not 40,000

The addressable pool of ex-TikTok Trust & Safety engineers, meaning people who actually shipped rule engines, ML classifiers, and moderation platforms inside ADSO, is roughly two orders of magnitude smaller than the "ex-TikTok T&S" haystack recruiters keep pulling from LinkedIn. In Refolk's index of professional profiles, only about 6 current or recent TikTok employees hold engineering-pattern titles that also contain "Trust and Safety," "Content Moderation," or "Integrity." The broader ex-TikTok/ByteDance T&S universe across all functions (policy, QA, vendor management, moderation) is 139.

The 40,000 figure comes from TikTok CEO Chew Shou Zi's January 2024 Senate Judiciary testimony, where he stated the company employed more than 40,000 professionals globally in trust and safety roles. That number is real. It is also almost entirely moderators, policy analysts, and vendor-managed contractors. Chew was not counting the people who wrote the classifier training pipelines.

800×

If you are staffing a T&S platform team in Q3 2026, treat this as a named-list exercise, not a boolean search. There are under 50 people in the world you actually want, and a meaningful share of them have competing internal TikTok offers.

What the July 2026 ADSO restructure actually freed up

The July 2026 wave is a consolidation of TikTok's ADSO (AI Data Service and Operations) model that moves platform IP into regional hubs while releasing the vendor-management and moderation layers underneath. Staff in Dublin were told the company is "considering a proposal to restructure the ADSO operating model in Ireland" to consolidate quality assurance into certain regional hubs where the work is already happening. The Journal reported the proposal covers roughly 300 of TikTok's more than 2,000 Irish staff, and most are in vendor management and content moderation.

Here is what the wave looks like when you stack it against the 2024 cuts:

The Irish Times noted TikTok is also proposing to create hundreds of new specialist T&S jobs in Dublin, with some affected staff likely moving to them. Every good candidate has a competing internal offer.

Why the Dublin geography is a trap

Dublin is where the layoff notices landed, not where the engineers are. In Refolk's index, only about 1.4% of the 139-person ex-TikTok T&S universe (2 people) sits in Greater Dublin. Malaysia carries about 2.9% (4 of 139), residue from the 500-person 2024 cut. The rest is distributed across Mountain View, Singapore, and Beijing, which is where the ADSO rule-engine and classifier work has historically clustered.

If you build a Dublin-anchored sourcing list, you get policy analysts, QA leads, and vendor managers. You do not get the ML platform engineer who wrote the takedown-latency dashboard.

The Berlin 150 are AI trainers, not AI builders

The Berlin strikers rallied under the banner "We trained your machines - pay us what we deserve." That slogan is a candidate-signal you should read literally. The Berlin 150 largely labelled data for the classifiers rather than shipping them, which means for Discord or Roblox they are an RLHF-labeling and red-team talent pool, not a platform-engineering pool. Different req. Different comp band.

Two other things follow from Berlin that shape how you approach every candidate from this wave:

• Severance anchors are unusually high. German union ver.di called the strike demanding severance equal to three years' salary per employee. Even where the demand did not land in full, it reset walk-away expectations across the broader affected population.
• The AI-training experience is real and rare. Candidates who spent months labeling adversarial content for a production classifier have exactly the intuition Character.AI and Roblox are hiring for right now. It is just not "engineer" work.

Every good ex-TikTok candidate from this wave has a competing internal offer and a three-years-salary severance anchor. Cycles will run long.

The dataset: what Refolk's index actually shows

The numbers below are the spine of the sourcing plan. Everything else is noise pulled from Chew's Senate figure.

The gap between "Trust and Safety Engineer" as an exact title (3 profiles globally) and the actual population of engineers doing this work is the entire reason exact-string boolean fails here. The people you want have titles like "Senior Software Engineer, Integrity Platform," "ML Engineer, Community Health," or "Platform Engineer, Content Understanding." No boolean captures that variance across ByteDance's naming conventions plus every adjacent employer.

This is the exact gap Refolk closes. You describe the person in plain English ("engineers who built content classifiers or moderation platforms at TikTok, ByteDance, Meta Integrity, or YouTube Trust & Safety") and get a ranked shortlist that treats the titles as fuzzy and the actual work as the signal.

Where the pool actually clusters (and who else is bidding)

The addressable engineers cluster in Mountain View, Singapore, and Beijing, and the competing bidders for them are a mix of large VLOPs and well-funded T&S platform vendors. Sourcing plans that only target consumer platforms will lose a meaningful share of the pool to the tooling companies.

Named competitors for the same pool:

• Consumer platforms hiring DSA-critical T&S engineers: Discord, Roblox, Character.AI, Bumble, Pinterest, Match Group.
• Infra and safety layers: Google Jigsaw, Cloudflare's T&S product, Thorn (CSAM classifiers).
• T&S platform vendors that explicitly raise on ex-TikTok/Meta hires: Hive AI, ActiveFence, Checkstep, Cinder.

The non-obvious bidders are the last group. Hive, ActiveFence, Checkstep, and Cinder sell the moderation platform back to Discord, Roblox, and the smaller VLOPs. Their pitch to candidates is "build the tool for the whole industry, not just one app." If you are Discord and you are competing for the same engineer, you are competing with the vendor Discord itself buys from.

DSA compliance is the wedge, not moderation

The engineers who built ADSO's audit trails, transparency-report tooling, and takedown-latency dashboards are the sub-pool DSA-regulated platforms will bid for hardest, and there are maybe 15 to 25 of them globally. The Digital Services Act (the EU regulation requiring very large online platforms to publish transparency reports, submit to audits, and evidence takedown timing) turned this specific skillset from cost-center engineering into an executive-visible line item.

Three signals that separate a DSA-relevant engineer from a general T&S engineer on a candidate profile:

1. Owned or contributed to a transparency-report pipeline. Look for mentions of Article 15/24 reporting, statement-of-reasons databases, or the EU's DSA Transparency Database submission API.
2. Shipped takedown-latency instrumentation. Any mention of SLA dashboards for content actions, appeal-turnaround metrics, or trusted-flagger workflows.
3. Audit-trail schema work. Engineers who designed the immutable-log side of moderation actions for external audit read very differently from engineers who trained classifiers.

Describe those three signals in plain English inside Refolk and you get a shortlist that is closer to 20 than 200. That is where DSA compliance engineering hiring actually happens.

How to run the sourcing motion in the next 60 days

Treat the July 2026 window as a two-month named-list exercise, not a pipeline build. The pool is small enough that every good candidate will hear from six recruiters, and severance anchors mean many will not respond to generic outreach quickly.

A concrete sequence:

1. Build the named list first, not the boolean. Start from the sub-50 addressable pool. Enrich each profile with their specific ADSO subsystem (rule engine, classifier, platform, transparency tooling).
2. Skip Dublin as a geography filter. Filter on employer and skill, not on layoff site. The Dublin cut generates news, not candidates.
3. Segment Berlin separately. The 150 Berlin strikers are AI-training and red-team talent. Route them to a different req with a different comp band and a different pitch.
4. Price against ver.di. Assume any German or Irish candidate has a walk-away number benchmarked against the three-years'-salary demand. If your offer does not clear a credible signing bonus, do not open the conversation.
5. Move on the vendor-platform bidders. If you are hiring for Discord or Roblox, your loss rate to Hive, ActiveFence, and Cinder will be higher than your loss rate to Meta. Reference-check accordingly.

The whole motion depends on collapsing the 40,000-profile haystack to the people who matter, fast enough that your top three targets have not already signed with Character.AI.

FAQ

How many ex-TikTok Trust & Safety engineers are actually on the market from the July 2026 cuts?

The realistic upper bound is under 50, and the tight core is smaller. Refolk's index shows only 6 current or recent TikTok employees with T&S/Content Moderation/Integrity in engineering-pattern titles, against a broader 139-person universe that includes all functions. The 40,000 figure Chew Shou Zi cited to the Senate in January 2024 is real but includes moderators, policy staff, and vendor-managed contractors. It is not an engineer count.

Should I source in Dublin because that is where the July 2026 layoffs hit?

No, and this is the single most common mistake on this wave. Only about 1.4% of the 139-person ex-TikTok T&S universe in Refolk's index sits in Greater Dublin. The engineering IP for ADSO's rule engines and classifiers has historically clustered in Mountain View, Singapore, and Beijing. Dublin will surface policy analysts, QA leads, and vendor managers, which are different roles at different comp bands.

Are the Berlin strikers a good fit for ML engineering roles at Discord or Roblox?

Mostly not for ML engineering, but they are excellent for adjacent roles. The Berlin 150 rallied under "We trained your machines," and their work was largely labeling adversarial content for classifiers rather than building the classifiers themselves. That makes them strong candidates for RLHF-labeling leads, red-team roles, and safety-policy operators at Character.AI, Discord, and Roblox. Different req, different comp band.

What is the DSA sub-pool and why should I prioritize it?

The DSA sub-pool is the 15 to 25 engineers globally who built transparency-report tooling, audit trails, and takedown-latency dashboards to satisfy the EU's Digital Services Act. Every very large online platform, plus smaller VLOPs like Bumble and Pinterest, needs this specific skillset, and it did not exist as a hiring category two years ago. If you are a DSA-regulated platform, this is the sub-pool where every hire pays for itself in reduced audit and compliance exposure.