Anthropic Fellows Program for AI Safety Research 2026 offers funding, mentorship and research opportunities for engineers and researchers interested in developing solutions to important AI safety challenges.
According to Anthropic’s Alignment Science Blog, fellows spend four months conducting empirical research aligned with Anthropic’s research priorities, working closely with Anthropic researchers. The programme is designed to help participants transition into empirical AI safety research and produce public research outputs, such as papers.
The programme welcomes candidates from a range of quantitative and technical backgrounds and does not require a PhD, previous machine-learning experience or published research papers.
Background & Programme Description
The Anthropic Fellows Program provides selected fellows with funding and mentorship to investigate high-priority questions in AI safety research.
Anthropic says that more than 80% of fellows in its first cohort produced papers, while more than 40% subsequently joined Anthropic full-time.
Recent cohorts have expanded into several areas, including:
- Scalable oversight
- Adversarial robustness
- AI control
- Model organisms
- Mechanistic interpretability
- AI security
- Model welfare
The programme is particularly aimed at people who want to transition into empirical research focused on reducing risks associated with advanced AI systems.
What Fellows Work On
Fellows work for four months on empirical research questions connected to Anthropic’s research priorities.
Anthropic mentors propose project ideas, after which fellows select and shape their projects in collaboration with their mentors.
The programme aims to produce public outputs, such as research papers.
Security
Security-focused projects explore ways AI systems could potentially be misused for cyberattacks and how researchers can develop rapid responses to emerging jailbreaks.
Anthropic reports that fellows have worked on AI-driven cybersecurity research, including identifying vulnerabilities in blockchain smart contracts and developing techniques for responding to high-risk jailbreaks.
Interpretability
Interpretability research focuses on understanding the internal workings of large language models.
Anthropic describes research involving attribution graphs, which can partially reveal the internal steps a model takes when producing an output.
The associated work has also been made publicly available to enable researchers to trace circuits, visualize and annotate graphs, and test hypotheses involving model features.
Model Organisms
Model-organism research uses controlled demonstrations of potential alignment failures to improve understanding of how these failures could emerge.
Anthropic describes fellows investigating agentic misalignment using simulated corporate environments and studying subliminal learning, in which behavioural traits can be transmitted through seemingly unrelated training data.
Funding and Benefits
According to the supplied programme information, fellows receive:
- USD $3,850 per week
- £2,310 per week
- CAD $4,300 per week
- Approximately $15,000 per month in compute funding
- Close mentorship from Anthropic researchers
The stipend is presented in different currencies in the source. Applicants should confirm the applicable payment terms and currency directly through the official application.
The programme also provides an opportunity to work closely with researchers in AI safety and develop practical research experience.
Qualifications
Education and Certification
A PhD is not required.
Anthropic specifically states that candidates do not need:
- A PhD
- Prior machine-learning experience
- Published research papers
Successful fellows have come from backgrounds including:
- Physics
- Mathematics
- Computer science
- Cybersecurity
- Other quantitative fields
This makes the programme potentially accessible to technically strong candidates who have not followed a traditional AI research career path.
Technical Skills
Strong candidates typically have good Python programming skills and the ability to work through ambiguous technical problems.
Applicants should be capable of:
- Coding effectively in Python
- Breaking ambiguous problems into concrete tasks
- Thinking clearly about difficult technical questions
- Learning new technical skills quickly
- Debugging problems
- Driving research projects forward despite uncertainty
Motivation
Anthropic is looking for people who are genuinely motivated by AI safety.
Candidates should be interested in reducing catastrophic risks from advanced AI systems and in transitioning into empirical AI safety research.
Experience
Prior machine-learning research experience is not mandatory according to the supplied information.
Instead, Anthropic places significant emphasis on a candidate’s ability to execute research, technical fundamentals, learning ability and motivation.
This means candidates from quantitative disciplines may be considered even if they have not previously worked professionally in AI safety.
Additional Information
Programme Duration
The fellowship lasts four months.
During this period, fellows work on empirical research projects with mentorship from Anthropic researchers.
Research Areas
The programme has recently expanded to cover a broad range of AI safety topics, including:
- Scalable oversight
- Adversarial robustness
- AI control
- Model organisms
- Mechanistic interpretability
- AI security
- Model welfare
The specific projects available can vary between cohorts.
Career Development
The programme can provide a pathway into longer-term AI safety research.
Anthropic reports that more than 40% of fellows from its first cohort subsequently joined Anthropic full-time, while others went on to work full-time on AI safety at other organisations.
This is an outcome reported by Anthropic and should not be interpreted as a guarantee of employment.
How to Apply
According to the supplied source, applications for the November 2026 cohort were scheduled to close on July 26, 2026.
Because the current date is August 19, 2026, that stated deadline has already passed.
Applicants interested in future cohorts should monitor the official Anthropic careers and Alignment Science pages for new application announcements.
Official Anthropic application information: Anthropic Fellows Program Application
Anthropic Alignment Science Blog: Alignment Science Blog
Who Should Consider This Programme?
The fellowship may be particularly relevant to:
- Python programmers
- Researchers interested in AI safety
- Quantitative researchers
- Computer science graduates
- Mathematics graduates
- Physics graduates
- Cybersecurity professionals
- Engineers
- Researchers transitioning into AI safety
- Candidates interested in AI alignment and interpretability
A traditional AI research background is not necessarily required, provided the candidate demonstrates strong technical ability and the capacity to conduct empirical research.
Frequently Asked Questions
Is a PhD required?
No. The supplied Anthropic information explicitly states that a PhD is not required.
Do applicants need previous machine-learning experience?
No. Anthropic says prior ML experience is not required.
Do I need published research papers?
No. Published papers are not listed as a requirement.
How long is the fellowship?
The fellowship lasts four months.
How much is the stipend?
The programme information lists a weekly stipend of $3,850 USD, £2,310 GBP or $4,300 CAD, plus approximately $15,000 per month in compute funding.
What research areas are covered?
Recent cohorts have worked across areas including scalable oversight, adversarial robustness, AI control, model organisms, mechanistic interpretability, AI security and model welfare.
Can people from non-AI backgrounds apply?
Yes. Anthropic reports successful fellows from physics, mathematics, computer science, cybersecurity and other quantitative backgrounds.
Is employment at Anthropic guaranteed after the fellowship?
No. Although Anthropic reports that more than 40% of fellows from its first cohort subsequently joined the company full-time, this is not a guarantee of employment.
Final Thoughts
The Anthropic Fellows Program for AI Safety Research is designed for technically capable researchers and engineers who want to transition into empirical AI safety research.
Its distinctive feature is that applicants are assessed much more on their ability to execute research, technical fundamentals, motivation and ability to learn than on conventional academic credentials.
The programme offers a four-month research experience, a weekly stipend, compute funding and close mentorship from Anthropic researchers. It also covers a broad range of research areas, from AI security and interpretability to scalable oversight and model organisms.
For the November 2026 cohort described in the supplied source, the application deadline was July 26, 2026, so that specific application window has passed. Interested candidates should watch Anthropic’s official channels for the next cohort.
SEO Keywords
Anthropic Fellows Program 2026, Anthropic Fellowship 2026, Anthropic AI Safety Fellowship, Anthropic Fellows Program AI safety research, AI safety research fellowship 2026, AI alignment fellowship, Anthropic research fellowship, AI safety jobs 2026, AI research fellowship, machine learning research fellowship, responsible AI research, AI safety opportunities, AI alignment research opportunities, Anthropic AI research, AI security research fellowship, mechanistic interpretability fellowship, AI safety careers, Python AI research fellowship, AI research opportunities 2026, Anthropic Fellows stipend.
