C8 — Where do MATS alumni go? And what do they attribute their outcomes to?

Context

MATS sent a Q3 2025 alumni survey with 123 respondents spanning cohorts 1.0–8.1 (heaviest representation in 6.0/7.0/8.0). The survey captures outcomes we can't see elsewhere: where alumni are working now, what mechanism they attribute MATS's impact to, whether they received post-MATS funding, and self-reported publication output.

This is the most downstream outcome data we have, and it comes with major caveats — it's an opt-in survey heavily skewed toward engaged/successful alumni, and self-attribution of MATS's causal contribution should be read as the alum's subjective view, not a controlled estimate.

MATS (Machine Alignment, Transparency & Security) is an AI safety research fellowship that places ~120 fellows with ~100 mentors per cohort. Cohort 10.0 ran in summer 2026 and was the first cohort with a centralized application review instead of decentralized stream-specific review. This analysis is part of a broader effort to evaluate the 10.0 process and inform the design of 11.0 (autumn 2026).

How the 10.0 selection pipeline worked (click to expand)

The 10.0 pipeline in brief. ~2,200 people applied. Each applicant went through three stages:

  1. Stage 1 — submitted background / experience / motivation, picked which research tracks they were interested in (Empirical, Policy & Strategy, Technical Governance, Theory, Compute Infrastructure). An LLM screen filtered out applicants who clearly didn't meet a minimum bar, and produced advisory per-stream recommendations.
  2. Stage 2 — applicants who passed Stage 1 had their materials scored by LLM-graded rubrics. The empirical track used a composite score combining Research Skills, Technical Execution (split into MLE, SWE, Math sub-scores), and Soft Skills. The top ~600 by composite advanced to Stage 3.
  3. Stage 3 — applicants chose specific mentors / "streams" to apply to. Each stream reviewed its applicants and produced a ranked list. Top-ranked applicants got offers; lower-ranked got waitlisted. ~120 offers were made.

For the empirical track, the composite formula is 0.50·RS + 0.35·TE + 0.15·SS, where TE = 0.50·MLE + 0.30·SWE + 0.20·Math. A "relevance multiplier" (Direct=1.0 / Adjacent=0.85 / Distant=0.60) is applied to Research Skills based on how the applicant's experience matches the streams they applied to.

Outcome definitions (click to expand)

Outcome definitions used throughout these analyses:

Key caveats

Outcomes by latest cohort

Latest cohort n P(any pub) P(top-tier conf) P(funding) P(AI safety org) P(frontier lab)
4.0 10 70% 30% 90% 90% 80%
5.0 6 100% 67% 83% 83% 50%
5.1 4 100% 75% 100% 100% 100%
6.0 11 64% 36% 45% 100% 91%
6.1 12 83% 50% 83% 100% 67%
7.0 10 70% 30% 30% 100% 80%
7.1 20 75% 30% 30% 90% 70%
8.0 15 40% 13% 20% 100% 87%
8.1 18 33% 22% 22% 94% 100%

Org types by cohort

Self-attributed MATS impact mechanisms (top 15)

Mechanism # respondents
Increased research/technical skill 32
Strengthened Resume 30
Professional connections from MATS 24
Reference from mentor 21
1-1 support from MATS team 13
Other 6

Takeaways

  1. Self-reported AI-safety-org affiliation is high — across cohorts, 50–80%+ of survey respondents report working at an AI safety non-profit or for-profit. Strong opt-in bias toward engaged alumni.
  2. Publication rates and funding rates are sizable — most respondents report at least one post-MATS publication and a meaningful share have received post-MATS funding. Again, biased upward by who chooses to fill out the survey.
  3. Most-attributed mechanisms are about connections and validation. "Professional connections from MATS" and "Strengthened resume" dominate the list. Direct skill development is named less often than network / credentialing effects.
  4. For 11.0: the survey is a useful complement to mentor evals / publications. Continue running it. For interpretation, treat it as evidence about the upper-bound of MATS impact — biased favorably but informative about what the program does well.
🔧 Debug — how the data was interpreted (click to expand; safe to skip)

Sample. Alumni survey respondents with apps-data-cohort latest attribution. n=106.

Outcome variable(s). Self-reported binaries: any post-MATS publication, top-tier conference paper, received post-MATS funding, current employer type (AI safety, frontier, government, academia).

Predictor fields. N/A — descriptive cross-cohort summary.

Filters applied. Latest-cohort attribution restricted to cohorts with apps data (4.0+).

Missing-data handling. NaN treated as negative (e.g., NaN funding = no funding reported).

Key assumptions / caveats.