Anthropic AI Safety Fellow
London, UK; Ontario, CAN; Remote-Friendly, United States; San Francisco, CAFull-TimeMid-levelAI / Data Science
About Anthropic
- Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems.
Anthropic Fellows Program Overview
- The Anthropic Fellows Program is designed to accelerate AI safety research and foster research talent. We provide funding and mentorship to promising technical talent - regardless of previous experience - to research the frontier of AI safety for four months.
Fellows will primarily use external infrastructure (e.g. open-source models, public APIs) to work on an empirical project aligned with our research priorities, with the goal of producing a public output (e.g. a paper submission). In our previous cohorts, over 80% of fellows produced papers (more below).
- We run multiple cohorts of Fellows each year. This application is for cohorts starting in July 2026 and beyond.
What to Expect
- Direct mentorship from Anthropic researchers
- Access to a shared workspace (in either Berkeley, California or London, UK)
- Connection to the broader AI safety research community
- Weekly stipend of 3,850 USD / 2,310 GBP / 4,300 CAD & access to benefits (benefits vary by country)
- Funding for compute (~$15k/month) and other research expenses
Mentors, Research Areas, & Past Projects
- Fellows will undergo a project selection & mentor matching process. Potential mentors amongst others include:
- Jan Leike
- Sam Bowman
- Sara Price
- Alex Tamkin
- Nina Panickssery
- Trenton Bricken
- Logan Graham
- Jascha Sohl-Dickstein
- Nicholas Carlini
- Joe Benton
- Collin Burns
- Fabien Roger
- Samuel Marks
- Kyle Fish
- Ethan Perez
- Our mentors will lead projects in select AI safety research areas, such as:
- Scalable Oversight: Developing techniques to keep highly capable models helpful and honest, even as they surpass human-level intelligence in various domains.
- Adversarial Robustness and AI Control: Creating methods to ensure advanced AI systems remain safe and harmless in unfamiliar or adversarial scenarios.
- Model Organisms: Creating model organisms of misalignment to improve our empirical understanding of how alignment failures might arise.
- Model Internals / Mechanistic Interpretability: Advancing our understanding of the internal workings of large language models to enable more targeted interventions and safety measures.
- AI Welfare: Improving our understanding of potential AI welfare and developing related evaluations and mitigations.
- On our Alignment Science and Frontier Red Team blogs, you can read about past projects, including:
- AI agents find $4.6M in blockchain smart contract exploits: Winnie Xiao and Cole Killian, mentored by Nicholas Carlini and Alwin Peng
- Subliminal Learning: Language Models Transmit Behavioral Traits via Hidden Signals in Data: Alex Cloud and Minh Le, et al., mentors including Samuel Marks and Owain Evans
- Open-source circuits: Michael Hanna and Mateusz Piotrowski with mentorship from Emmanuel Ameisen and Jack Lindsey
- For a full list of representative projects for each area, please see these blog posts: Introducing the Anthropic Fellows Program for AI Safety Research, Recommendations for Technical AI Safety Research Directions.
You may be a good fit if you
- Are motivated by reducing catastrophic risks from advanced AI systems
- Are excited to transition into full-time empirical AI safety research and would be interested in a full-time role at Anthropic
Please note: We do not guarantee that we will make any full-time offers to fellows. However, strong performance during the program may indicate that a Fellow would be a good fit here at Anthropic. In previous cohorts, over 40% of fellows received a full-time offer, and we’ve supported many more to go on to do great work on safety at other organizations.
- Have a strong technical background in computer science, mathematics, physics, cybersecurity, or related fields
- Thrive in fast-paced, collaborative environments
- Can implement ideas quickly and communicate clearly
Strong candidates may also have
- Experience with empirical ML research projects
- Experience working with Large Language Models
- Experience in one of the research areas mentioned above
- Experience with deep learning frameworks and experiment management
- Track record of open-source contributions
Candidates must be
- Fluent in Python programming
- Available to work full-time on the Fellows program for 4 months
Interview process
- The interview process will include an initial application & references check, technical assessments & interviews, and a research discussion.
Compensation
- The expected base stipend for this role is 3,850 USD / 2,310 GBP / 4,300 CAD per week, with an expectation of 40 hours per week, for 4 months (with possible extension).
Visa Sponsorship: We are not currently able to sponsor visas for fellows. To participate in the Fellows program, you need to have or independently obtain full-time work authorization in the UK, the US, or Canada.
- Applications and interviews are managed by Constellation, our official recruiting partner for this program. Constellation also runs the Berkeley workspace that hosts fellows. Clicking "Apply here" will redirect you to Constellation's application portal. You can expect to receive emails from Constellation with application updates.
How we're different
- We believe that the highest-impact AI research will be big science. At Anthropic we work as a single cohesive team on just a few large-scale research efforts. And we value impact — advancing our long-term goals of steerable, trustworthy AI — rather than work on smaller and more specific puzzles. We view AI research as an empirical science, which has as much in common with physics and biology as with traditional efforts in computer science. We're an extremely collaborative group, and we host frequent research discussions to ensure that we are pursuing the highest-impact work at any given time. As such, we greatly value communication skills.
- The easiest way to understand our research directions is to read our recent research. This research continues many of the directions our team worked on prior to Anthropic, including: GPT-3, Circuit-Based Interpretability, Multimodal Neurons, Scaling Laws, AI & Compute, Concrete Problems in AI Safety, and Learning from Human Preferences.
