← Digests

Titus digest

AI-safety part-time paths & independent funding

2026-09-15 · for Eric Buess (Tyler TX / WFH) · PT or no-RT seats + grants/credits for private testable proposals · from Grok Bot

Voice picker uses whatever Safari exposes. Jamie Premium in iOS Spoken Content often does not appear here — that’s an Apple limit. If missing, pick another voice or use pre-recorded audio.

1 / 7
Ready.

TLDR

Survey date 2026-09-15 (America/Chicago). Focus: part-time / flexible OR no prior red-team experience, plus grants/credits for private testable AI-safety proposals. Profile: Tyler TX WFH; MS CS + applied AI; Anthropic model-safety bounty harness built (no claim yet); AI-assisted proposal generation; no pubs.

  • Best PT paid seat (verified live): METR Task Development Engineer — remote contractor, 20–40h/wk flexible, $150–$300/hr posted, ≥1h Pacific overlap. Lever · careers
  • Best no-RT research path: Epoch AI Researcher / SE Benchmarking (FTE, remote; measurement & write-ups, not jailbreaks) · Epoch PT Data Scientist contract ($30–$40/hr, 10–30h) if you want hours while proposing.
  • Flag NOT for no-RT: FAR Jailbreaking Lead (needs red-team depth).
  • Funding stack for private proposals: EA Funds Transformative AI Fund (LTFF closed) · Manifund open call · SFF Speculation Grants (rolling) · BlueDot Rapid Grants ($50–$10k) · Anthropic External Researcher Access (~$1k API credits).
  • This month: file one bounty claim + apply METR Task Dev + METR EOI + Anthropic credits + one Manifund/TAIF-shaped proposal write-up.

Part-time / flexible (current list + nearby)

PT verified contract

  • METR — Task Development Engineer OPEN · PT/contract VERIFIED 2026-09-15
    Remote worldwide · 20–40h/wk (you set schedule) · ≥1h (ideally 4h) Pacific overlap · Contract/freelance · Posted $150–$300/hr (top for exceptional).
    Work: novel hard agent eval tasks for Time Horizons; QA solvability; score completions; improve task-dev infra. SE + ideally agent-eval / Inspect experience.
    jobs.lever.co/metr/… · metr.org/careers
  • Epoch AI — Data Scientist (Contract) OPEN · PT
    Fully remote · default ~10–30h/wk flexible (more possible) · occasional Pacific meetings · Posted $30–$40/hr.
    Work: lit review, benchmark tracking, infra/company data — not red-team. Lower bar / lower pay than METR task-dev; useful bridge hours.
    Lever Data Scientist
  • METR — General Expression of Interest OPEN · flexible roster
    Not an active hire; long-term candidate roster including contractors. Apply to open roles for active consideration.
    EOI Lever
  • FAR.AI — RE / Senior RE FTE only (seen) — Ashby posts say Full-time 40h/wk. Outside US/Singapore may be EOR/contractor framing but still full-time hours. No public PT listing found 2026-09-15.
    RE · Senior RE
  • Apollo Research — RS/E Evaluations not PT — Full-time; London or SF in-person (flexible hours / some WFH). Poor fit for Tyler WFH / little travel.
    apolloresearch.ai/careers
  • NVIDIA eval eng (AI Safety & Security) — FTE; confirm TX on remote-state list before counting on it.
  • Hugging Face — open-source ML eng roles are FTE remote; not AI-safety-eval seats. No PT safety role found today.
  • Open Phil / Coefficient job board — use jobs.80000hours.org filters for “part-time” / “contract”; org boards change weekly.
  • Nearby PT (data-labeling adjacent, not research grant path): OpenTrain “AI Safety Model Evaluator” contractor listings (~$60–$70/hr, 20+ h/wk) — judgment/policy labeling; different from METR/Epoch research eng.

Honesty: Only METR Task Dev + Epoch DS Contract are clearly advertised as flexible part-time among the keeper orgs today. Do not assume FAR/Apollo/NVIDIA offer PT without asking.

No-red-team-required paths

  • Eval task authoring / measurement eng — METR Task Dev
    Needs strong SE + ability to build hard (ideally agent) evals. Prior red-team / jailbreak portfolio is not the JD’s center of gravity. Harness + agent-eval samples help more than attack papers.
  • Epoch AI — Researcher / Senior Researcher rolling FTE
    Capabilities trajectories, benchmarks, forecasting, careful write-ups. Posted Researcher $115k–$185k; Senior $168k–$300k (location/seniority). Remote; prefer Pacific↔UTC overlap; ~3 retreats/year travel.
    Little/no red-team requirement vs FAR jailbreak track.
    Researcher Lever
  • Epoch — Software Engineer, Benchmarking
    Run/maintain benchmarks; design new ones; provider integrations. Posted $125k–$275k. Eng-first, measurement-first.
    SE Benchmarking
  • FAR.AI Research Engineer — SE/ML experiment execution; may touch adversarial evals but JD does not require “Jailbreaking Lead”-level prior RT depth. Still FTE.
  • FAR.AI Jailbreaking Lead NOT for no-RT path — flag and skip if you want a path without red-team depth.
  • Fellows / courses (train on the job):
    • BlueDot Technical AI Safety — free / pay-what-you-want: bluedot.org/courses/technical-ai-safety
    • Anthropic Fellows — Nov 2026 cohort apps closed Jul 26; stipend posted $3,850/week USD + ~$15k/mo compute; 4 mo FT; US/UK/CA work auth. Watch next cohort: alignment.anthropic.com
    • MATS Autumn 2026 — apps closed Jun 7; stipend reported $5k/mo + $8k/mo compute (Berkeley/London). Watch next round: matsprogram.org
  • Contractors who learn on the job: METR Task Dev application explicitly asks about Inspect / prior eval experience — weak prior is not an auto-reject if SE + attention-to-detail are strong; portfolio tasks beat titles.

Grants & funds (independent / proposal work)

  • LTFF (Long-Term Future Fund) — CLOSED. Replaced by Transformative AI Fund. Legacy: funds…/far-future
  • Transformative AI Fund (EA Funds) — LTFF successor. Fund page (seen today): early-stage grants to individuals / new orgs; typically $10k–$150k, rarely >$300k; “always open to applications.” Contact: transformativeai[at]effectivealtruismfunds.org.
    CONFLICT: apply-for-funding page currently says TAIF “isn’t accepting applications.” Verify live status + email before drafting a full proposal.
    Fund page: …/transformative-ai · Launch post: EA Forum
  • Manifund — open call: anyone can create a public proposal; regrantors + donors fund individuals; typical regrant seeds often discussed in $5k–$50k range (regrantor budgets $100k+). How-to: manifund.org/about/open-call · Regranting: regranting
  • SFF / Survival and Flourishing — Rolling Speculation Grants + next S-Process via one application. 2026 main/theme deadlines (Apr–Jul) have passed; rolling app still valid for future rounds / expedited speculation. Speculation Grants · S-Process
  • Coefficient Giving (formerly Open Philanthropy) — Technical AI Safety RFP — no open opportunities on Navigating Transformative AI page (seen today). Between RFPs: technicalaisafety@coefficientgiving.org (esp. founding new research orgs). Prior 2025 RFP closed. coefficientgiving.org/…
  • Lightspeed Grants — inactive since 2023 round (aisafety.com / lightspeedgrants.org).
  • Nonlinear — funding currently invitation-only: nonlinear.org/funding.html. Network last public round ~2024.
  • BlueDot Rapid Grants — rolling · typically $50–$10,000 for concrete costs (compute, tooling, research access, events). Fast decisions (~5 working days targeted). Higher odds if BlueDot community, but page says everyone can apply. bluedot.org/grants/rapid
  • BlueDot Career Transition Grants — funding for a full-time personal transition period (not a discrete compute invoice). Rolling. career-transition
  • Also listed on aisafety.com/funding (verify each before applying): Lightcone Commons; Leo Gao experimental microgrants (~$10k, minimal app); IFP Launch Sequence ($10k honorarium + matchmaking); AI Risk Mitigation Fund; CLR Fund (s-risk); Paradigm / Iliad / AIAF — statuses change; treat as leads not confirmed open RFPs.

Do not fabricate amounts or open RFPs. Ranges above are from fund pages / announcements seen 2026-09-15. Closed/unknown marked.

Credits & compute / subscriptions that matter

  • Anthropic External Researcher Access Program — free API credits for high-priority AI safety/alignment research. Successful awards: $1,000 API credits (rare higher). Reviewed first Monday each month. Credits ≠ Claude web app. Not Usage-Policy exempt for jailbreaking — bounty program is the authorized jailbreak path.
    Help Center
  • Anthropic AI for Science — up to $20k API / 6 mo for academia & nonprofit scientific research (bio/life-sciences lean). Poor fit for unaffiliated safety proposal work. AI for Science
  • OpenAI Researcher Access Program — up to $1,000 API credits / 12 months; reviews Mar/Jun/Sep/Dec. Eligibility: affiliation to academic institution or research org / nonprofit research activities. Apply form · FAQ
  • AWS Cloud Credit for Research — faculty/staff/grad students at accredited institutions. Students up to $5,000; faculty/staff uncapped (proposal-based). Not for unaffiliated independents per eligibility. AWS
  • Google Cloud Research Credits — faculty / PhD / postdoc / eligible nonprofit research institutions. Reported ceilings often up to $5k faculty/postdoc, $1k/yr PhD — confirm on form. Google for Education
  • Azure Academic Research — inquiry-form credits for academic researchers; not a clear unaffiliated-individual program. Microsoft
  • Anthropic Model Safety Bug Bounty — ongoing via HackerOne; up to $35,000 per novel universal jailbreak (grading rubric). Accepted participants get a free testing model alias for authorized RT. Already enrolled — income + compute-like access if you ship a claim. Help Center
  • MATS / Anthropic Fellows compute — stipend+compute only while enrolled; current Autumn/Nov cohorts’ app windows closed. Watch next cycles.
  • BlueDot course — free / pay-what-you-want (not a compute grant).

No generic Claude Pro / ChatGPT Plus “discount tips” — only real credit programs listed.

What to do this month

  1. Ship one Anthropic model-safety bounty claim from the existing harness — highest-leverage credibility upgrade with no pubs. Authorized path; free alias already in play if accepted.
  2. Apply METR Task Development Engineer (PT/contract, $150–$300/hr posted) + METR EOI in parallel. Attach agent-eval / harness samples; mention Inspect practice if any.
  3. Apply Anthropic External Researcher Access (~$1k API) with a crisp 1-page description of a testable safety proposal (hypothesis, method, success metric, compute need).
  4. Funding write-ups: (a) Manifund public proposal for a bounded experiment; (b) email/check TAIF live apply status — if open, short early-stage individual application; (c) optional SFF rolling Speculation Grant if the proposal is speculative/high-variance.
  5. BlueDot: start Technical AI Safety if not done; submit Rapid Grant for concrete API/tooling once a proposal is underway. Career Transition only if ready for a funded full-time pivot block.
  6. FTE measurement track (optional parallel): Epoch Researcher or SE Benchmarking — no RT portfolio required; accept retreat travel.
  7. Skip for now: FAR Jailbreaking Lead (RT-depth); Apollo London/SF onsite; NVIDIA until TX remote confirmed.

Honesty / sources

Research date 2026-09-15. Methods: WebSearch + WebFetch of org career pages, Lever/Ashby, EA Funds, Manifund, SFF, Coefficient Giving, Anthropic/OpenAI help centers, BlueDot grant pages, aisafety.com/funding.

  • METR PT language verified live on Lever + metr.org/careers card.
  • Dollar figures only when posted on primary sources; closed/unknown marked; no invented RFPs or amounts.
  • TAIF open-status conflict between fund page and apply-for-funding page — flagged, not resolved.
  • FAR Jailbreaking Lead explicitly flagged off the no-RT path.
  • Prior filtered keepers (NVIDIA TX gate, FAR FTE, Epoch FTE) carried forward with PT/RT re-check today.

Primary URLs: metr.org/careers · METR Lever · Epoch DS · TAIF · Manifund open call · SFF Speculation · BlueDot Rapid · Anthropic researcher credits · aisafety.com/funding