Is this opportunity actionable now?

closed. Four-month cohorts; the May/July 2026 call is closed and no next cohort is posted.

Which jurisdictions and applicants are eligible?

Candidates must be able to execute empirical research and code in Python; PhD, prior ML work and publications are not required. Work-location and authorization constraints must be taken from the live application, not inferred from the blog post.

Strong Python and technical fundamentals; Ability to turn ambiguous safety questions into concrete progress; Motivation to reduce catastrophic AI risk and evidence of learning/shipping

What burden or follow-on should I expect?

Full-time four-month mentored research aimed at public output such as a paper, with close project collaboration and substantial compute. No automatic renewal; strong fellows may be hired, but employment is not promised.

Anthropic Fellows Program (AI Safety)

$3,850 USD, £2,310 GBP or $4,300 CAD per week, plus approximately $15,000 per month of compute; the official source does not publish a guaranteed total stipend · Anthropic

Four months of mentored, empirical AI safety research (interpretability, scalable oversight, adversarial robustness/AI control, model organisms, AI security, model welfare) with close mentorship from Anthropic researchers. Current lifecycle: closed. Real economics: $3,850 USD, £2,310 GBP or $4,300 CAD per week, plus approximately $15,000 per month of compute; the official source does not publish a guaranteed total stipend.

Tier 3aifellowshipai-safetyresearch

Apply Official site Open in the live directory

Key facts

Funder	Anthropic
Amount	$3,850 USD, £2,310 GBP or $4,300 CAD per week, plus approximately $15,000 per month of compute; the official source does not publish a guaranteed total stipend
Decision time	Application screening and technical selection; the public post does not promise a fixed response time.
Eligibility	Candidates must be able to execute empirical research and code in Python; PhD, prior ML work and publications are not required. Work-location and authorization constraints must be taken from the live application, not inferred from the blog post.
Restrictions	Full-time four-month mentored research aimed at public output such as a paper, with close project collaboration and substantial compute. Dilution/equity: none.
Status	closed

Always confirm on the official page before applying — dates and terms change.

Application process

Apply to a published cohort
Technical and motivation screening
Anthropic mentors pitch project ideas
Fellow and mentor shape an empirical project
Execute for four months and aim for a public research output

Background

Anthropic operates Anthropic Fellows Program (AI Safety). The official-source pass separates the currently actionable opportunity from historical cohorts and separates per-recipient value from headline program envelopes. Anthropic says over 80% of first-cohort fellows produced papers and over 40% later joined Anthropic full-time.

How the application really works

Apply to a published cohort → Technical and motivation screening → Anthropic mentors pitch project ideas → Fellow and mentor shape an empirical project → Execute for four months and aim for a public research output

Tips

Demonstrate: Strong Python and technical fundamentals
Demonstrate: Ability to turn ambiguous safety questions into concrete progress
Demonstrate: Motivation to reduce catastrophic AI risk and evidence of learning/shipping

Watch out for

Current lifecycle is closed.
Dilution/equity treatment: none.
Full-time four-month mentored research aimed at public output such as a paper, with close project collaboration and substantial compute.

Track record

Anthropic says over 80% of first-cohort fellows produced papers and over 40% later joined Anthropic full-time.

No applicant count or acceptance rate published.

Frequently asked questions

Is this opportunity actionable now?: closed. Four-month cohorts; the May/July 2026 call is closed and no next cohort is posted.
What is the real per-recipient economic value?: $3,850 USD, £2,310 GBP or $4,300 CAD per week, plus approximately $15,000 per month of compute; the official source does not publish a guaranteed total stipend
Which jurisdictions and applicants are eligible?: Candidates must be able to execute empirical research and code in Python; PhD, prior ML work and publications are not required. Work-location and authorization constraints must be taken from the live application, not inferred from the blog post.
How does selection work?: Apply to a published cohort; Technical and motivation screening; Anthropic mentors pitch project ideas; Fellow and mentor shape an empirical project; Execute for four months and aim for a public research output. Application screening and technical selection; the public post does not promise a fixed response time.
What appears to win?: Strong Python and technical fundamentals; Ability to turn ambiguous safety questions into concrete progress; Motivation to reduce catastrophic AI risk and evidence of learning/shipping
What burden or follow-on should I expect?: Full-time four-month mentored research aimed at public output such as a paper, with close project collaboration and substantial compute. No automatic renewal; strong fellows may be hired, but employment is not promised.

Sources

alignment.anthropic.com/2025/anthropic-fellows-program-2026 — Official weekly stipend rates, monthly compute support, cohort length, eligibility and outcomes (checked 2026-08-10)