Anthropic Opens Applications for People Transitioning into AI Safety: Up to $500,000 a Year, No Machine Learning Background Required

Anthropic has opened applications for the Anthropic Residency. It is a paid, fixed-term, full-time research program offered in three tracks: six weeks, three months, and six months. It is aimed at people who have already made their mark in another technical field and want to transition into full-time AI safety research. The official listing explicitly states that a machine learning background is not required.

First, the hard facts. U.S. roles pay $350,000 to $500,000 per year, and U.K. roles pay £265,000 to £375,000 per year, prorated to the actual duration. For a full six months, the U.S. role pays $175,000 to $250,000. The deadline is 11:59 p.m. Pacific Time on November 22, which is 3:59 p.m. Beijing Time on November 23.

[Who should apply]

The official listing names three types of people: professors in quantitative disciplines, technical founders, and Staff-level engineers. Staff engineer is the level above senior engineer at large tech companies, the layer that can lead large projects. The listing states that a PhD, machine learning experience, or publication record is not required. Nice-to-haves include having written papers on machine learning, natural language processing, or AI safety, or having experience with large models, reinforcement learning, Kubernetes clusters, and large shared codebases.

There is something easy to miss here. Not requiring a machine learning background does not mean the bar is low. Someone in the comments said they got excited when they saw that a machine learning background was not required, but then saw that the roles were for professors, technical founders, and Staff engineers, and was immediately discouraged. That reaction is quite accurate. The program screens for people who have already demonstrated research judgment in another field. Career changers starting from zero are not in scope.

Anthropic's rationale is that machine learning is still young, and many senior researchers were originally physicists, mathematicians, neuroscientists, or software engineers. The stronger models become, the more important it is in research to set direction and judge what is worth doing. Senior people from other fields are often already good at this; what they lack is knowledge of frontier AI.

[What you will do]

Each person is paired with an Anthropic researcher as a mentor, with regular one-on-ones. Projects are proposed by Anthropic researchers and matched to individual interests and backgrounds. The program sets aside time for self-study and provides lectures and tutoring to help people build foundations in modern machine learning and AI safety. Participants can use frontier models, and projects have dedicated compute budgets.

Research directions fall into three areas. AI safety and alignment, including how to supervise models once they surpass humans, opening up models to see how they work internally, and AI welfare. AI security offense and defense and frontier red teaming, mapping out what models can do in areas such as cybersecurity, robotics, and cryptography. Model training and infrastructure, from model architecture, optimizers, and training data to accelerator simulators, hardware backends, and reinforcement learning training environments.

The official listing names nine possible mentors, including Ethan Perez, head of the alignment science team; Jack Lindsey, head of the model psychology team; Jascha Sohl-Dickstein, former principal scientist at Google Brain; and Nicholas Carlini, who researches adversarial machine learning and AI safety.

This program is modeled on Anthropic Fellows. Fellows is aimed at researchers new to the field and has produced research such as 'subliminal learning,' the Petri model auditing tool, and persona vectors. Residency applies the same model to seasoned technical leaders.

[Compensation and location]

U.S. roles pay $350,000 to $500,000 per year, and U.K. roles pay £265,000 to £375,000 per year, prorated to the actual duration, plus benefits. Anthropic would like people to work in the San Francisco office, but remote work is possible, especially for short-term projects. Remote participants must be located in countries where Anthropic can legally hire; the official listing does not list the specific countries. Visa sponsorship is available for three months and longer, and remote participation is encouraged for the six-week track. U.K. roles are remote in the U.K. by default, with the option to go to the London office.

Anthropic says the goal is for everyone to be able to continue doing AI research full-time after the program, whether at Anthropic or elsewhere. It also notes that most full-time research roles at Anthropic require relocating to San Francisco.

[How to apply]

The job page has two entry points: U.S. roles and U.K. roles. The process is: submit an application, recruiter screen, technical assessment, research discussion, and background check, with offers expected in January 2027. Applicants may also be considered for the Fellows program at the same time.

Program overview: https://alignment.anthropic.com/2026/anthropic-residency/
U.S. role application: https://job-boards.greenhouse.io/anthropic/jobs/5438311008
U.K. role application: https://job-boards.greenhouse.io/anthropic/jobs/5439169008
Official tweet: https://x.com/AnthropicAI/status/2117343929960112486

The replies under the official tweet are also pretty direct. Someone posted a video of a cat wearing a green scarf staring at a laptop, captioned "on it." Someone else saw the $350,000 salary and posted a screenshot from a comedy show. Another joked about applying to protect Claude and having Claude write the cover letter.

One commenter noted that the six-week track is barely enough to catch up on fundamentals, and is more like a taste test before deciding whether to actually make the switch. That judgment makes sense. Six weeks is suitable for testing the waters; three months and longer is where visa sponsorship and a complete project become possible.

What is worth paying attention to in this program is the screening criteria. The salary is only the surface; the standard for hiring in AI safety is changing. Publication count still matters, but there are now multiple paths in. For ordinary readers, the signal in this news is also clear: AI safety is starting to reach beyond pure machine learning backgrounds toward senior experience in other fields. For AI practitioners with development backgrounds, the Staff engineer path is a real opportunity; for people who have not reached that level, the six-week track is more of a test run—a way to see whether they really want to make the switch.