Anthropic opens research residency paying half a million dollars annually and removes machine-learning background requirement

Anthropic’s lab opened applications for the new research residency program aimed at academics, technical founders and senior engineers, in a move that removes the requirement for prior machine-learning experience to work on transformative AI.The program provides a financial stipend ranging from $350,000 to $500,000 per year, and runs over fixed periods that span from six weeks to six months.The application deadline for this round is November 22 2026, with flexible start dates and the possibility of remote participation, especially for short terms, although the lab prefers work from its headquarters in San Francisco.

The program clearly targets university professors in quantitative fields, as well as technical founders and senior-level engineers who wish to direct their expertise toward AI safety and making it beneficial for everyone. Anthropic gives accepted participants direct access to advanced frontier models, allocates a defined compute budget for their research projects, and pairs each researcher with a dedicated mentor from within the lab who provides individual, regular guidance throughout the residency. The application format lets participants choose their preferred term, whether six weeks, three months or a full six months.

The list of mentors available in the program reveals the range of expertise needed to address advanced model-safety challenges, and explains why the application is no longer limited to specialists in traditional machine learning.The mentorship roster includes research leaders specialized in mathematics, neuroscience, bioinformatics and cybersecurity.Among them is Samuel Marks, who leads the scalable supervision team with a background in mathematics; Ilana Simon, a researcher in interpretability of language and biological models from a biomedical informatics perspective; Jack Lindsey, head of the upcoming model psychology team from theoretical neuroscience; Nicolas Carlin, a researcher in adversarial machine learning and AI safety coming from DeepMind and Google Brain; Yasha Sol-Dextein, who works at the intersection of safety and pre-training with a background that combines physics and neuroscience; Sara Price, a leader in alignment training; Ethan Perez, head of alignment science; Evan Hopinger, leader of alignment stress testing from the Machine Intelligence Research Institute; and Belinda Lee, who specializes in how models represent the world, users and themselves.

This shift sends a direct practical signal to tech markets and universities in the Gulf, Egypt and the Levant, where institutions have assumed that moving to advanced model research requires years of academic re-training exclusively within software-engineering and machine-learning tracks. The program creates a tangible pathway for academic talent in science and mathematics faculties and seasoned engineers in regional tech hubs to engage in global alignment research without abandoning their original scholarly credentials, with the option of remote participation for short terms.Changing the competency requirements at frontier labs means that the talent-attraction battle is no longer about writing abstract code, but about the ability to tune systems and dissect model behavior mathematically and behaviorally.This development puts new pressure on technical team leaders and hiring managers in Riyadh, Abu Dhabi and Cairo to revisit competency evaluation criteria, as competition for mathematical and systems talent has become global and commands salaries that exceed traditional hiring packages.

The launch of the residency reflects the fact that model-safety challenges and control over outputs have moved beyond superficial statistical training to deeper problems that require rigorous experimental and mathematical thinking. With the application deadline approaching on November 22 2026, Anthropic offers a practical model that shows closing the safety gap in intelligent systems involves opening labs to experts with deep quantitative expertise and providing them with appropriate compute infrastructure, rather than waiting for the traditional re-training of the job market.