Anthropic Residency: applications open for established technical leaders

The autumn 2026 announcement of the Anthropic Residency, a paid fixed-term research program of six weeks, three months or six months for professors in quantitative fields, technical founders and staff-level engineers moving into AI safety research, paid at $350,000 to $500,000 a year equivalent, with nine named potential mentors, research areas from alignment to ML systems, and applications closing on 22 November 2026.

Anthropic (Alignment Science Blog) · 2026-11-02 · original

The Anthropic Residency is a paid, fixed-term research program for people who have already led technical work in another field and want to move into AI safety research. This Alignment Science Blog post opened applications for it. The blog's index lists the post under October 2026; the earliest records found are a Wayback Machine capture and a news report both from 2 November 2026, so this note gives 2 November as the latest possible date and the index's October as the blog's own month. The post reports no research results. Its argument for the program is that "what matters most in AI research is increasingly the ability to frame a research agenda and judge what is worth doing", skills it says senior people in quantitative fields often already have.

Who the Residency is for

The post names three groups: a "professor in a quantitative discipline, a technical founder, or a staff-level engineer". It states that a background in machine learning is not required, and that applicants do not need a PhD, prior ML experience or papers. Applicants to the Residency "may also be considered" for the Anthropic Fellows Program.

Terms as stated

The section "What to expect in this round" sets out the terms:

  • Length. Placements of six weeks, three months or six months, with flexible start dates.
  • Pay. "$350,000 to $500,000/year equivalent".
  • Work. Problems outlined by Anthropic researchers, matched to the resident's interests; each resident is paired with an Anthropic mentor for regular one-on-one guidance.
  • Training. Dedicated self-study time, talks and guidance on modern ML and AI safety research.
  • Compute and models. Residents "will work with frontier models" and have a dedicated compute budget.
  • Place. San Francisco preferred, remote possible "especially for shorter placements"; the post notes that most full-time research roles at Anthropic require relocation to San Francisco.
  • Deadline. Applications close on 22 November 2026, through separate US and UK postings.

The post's stated aim is that every resident ends "in a position to continue working on AI research full-time, whether at Anthropic or elsewhere". It makes no promise of employment.

How it differs from the Fellows Program

The post says the round "builds on" the Fellows Program, "which demonstrated how quickly new researchers can contribute to the field", and adapts that model for established leaders. The differences in the stated terms are these (the comparison is this note's, from the two programs' posts): the Fellows Program, as described in its later announcement, runs four months with a weekly stipend set in a job description; the first Fellows announcement restricted projects to public APIs and open-source models (the Fellows job description, retrieved 13 November 2026, says fellows "primarily use external infrastructure"), whereas residents are to work with frontier models; and the Residency adds self-study and talks for people without ML training.

Mentors and research areas

The post lists nine potential mentors with their roles as it gives them: Sara Price (Alignment Training team lead), Ethan Perez (Alignment Science team lead), Evan Hubinger (Alignment Stress-Testing team lead), Samuel Marks (Scalable Oversight team lead), Elana Simon, Belinda Li, Jack Lindsey (Model Psychology team lead), Nicholas Carlini and Jascha Sohl-Dickstein. Its areas fall under three headings: AI safety and alignment (scalable oversight, adversarial robustness and AI control, model organisms, model internals and interpretability, safeguards research, AI welfare); AI security and frontier red-teaming; and ML systems and reinforcement learning, which covers how frontier models are trained (architecture, optimizers, training data, simulators, RL environments). The last heading concerns training frontier models rather than safety; the Fellows job description (retrieved 13 November 2026) lists safety and security workstreams but also mentions "fellows on capabilities-focused projects".

Fellows work the post cites

The post names Fellows work as examples, each with a note here: Subliminal Learning, Petri and Persona Vectors in its text, and in its footnotes Automated Alignment Researchers, SHADE-Arena, SLEIGHT-Bench, AuditBench, Natural Language Autoencoders, Poisoning Fine-tuning Datasets of Constitutional Classifiers, AI agents find $4.6M in blockchain smart contract exploits and How Claude performs on robotics tasks. The footnotes label these "Representative research from the Anthropic Fellows Program". For three of them the bylines in the collection's notes name no Fellows affiliation, and the sources differ: the Natural Language Autoencoders paper thanks the program in its acknowledgments, saying Kit Fraser-Taliente was a member "when he invented the original NLA methodology"; SHADE-Arena's author contributions (arXiv:2506.15740v2, Appendix A) say Ethan Perez oversaw it as lead mentor "in the MATS stream and the Anthropic Fellows Program"; the robotics post, by the Frontier Red Team, does not mention the program, so for it the label is this post's alone.

Limitations and open questions

The post is a recruiting announcement for one round ("this round"); it does not say whether later rounds will follow. It does not say how many residents will be accepted, how they will be selected beyond the listed qualities, who owns their work, whether they are employees, or what they may publish. It reports nothing yet about outcomes.

Research questions

  1. How many residents were placed, and for how long? A later report from the program would answer it.
  2. Did residents publish, and in which areas? Bylines crediting the Residency in later Anthropic work would show its output and whether the ML-systems heading drew residents away from safety work.
  3. How many residents continued in AI research afterward? The post's stated aim makes this its own measure of success.
  4. Does working with frontier models change what residents can study, compared with Fellows? Comparing the models used in Residency and Fellows papers would show it.

How it connects

References