Anthropic Pays $500K to Lure Professors and Founders Into AI Safety

Anthropic has opened applications for the 2026 Anthropic Residency, a paid, fixed-term program that brings established technical leaders into full-time AI safety and alignment research. It targets tenured professors, startup founders, staff-level and more senior engineers, and other experienced quantitative researchers.

Machine-learning experience is optional, and residents receive time and support to learn current ML methods while applying expertise from other fields. Alignment research examines how to make advanced AI systems behave as intended and remain under effective human control as their capabilities improve.

Six weeks, six months, or somewhere between

Placements run from six weeks to six months, with applicants stating their preferred duration. Anthropic designed the range for candidates who may need to arrange leave from senior academic or industry roles.

• Schedule: Full time for the selected placement.
• Compensation: An annualized equivalent of $350,000 to $500,000.
• Mentorship: Regular one-on-one guidance from an Anthropic researcher.
• Training: Dedicated self-study time, talks, and guidance on modern ML and AI safety.
• Resources: A dedicated compute budget and access to frontier models, a term for the most capable current systems.
• Location: San Francisco is preferred, while remote participation may be possible, particularly for shorter placements.

Anthropic’s case for transferable judgment

Anthropic’s premise is that senior researchers can transfer skills in choosing useful questions, designing rigorous studies, and assessing evidence under uncertainty. Many researchers at the company began in physics, mathematics, neuroscience, or software engineering before moving into machine learning.

The residency extends the mentored format of the earlier Anthropic Fellows Program, whose participants contributed to research on subliminal learning, the open-source Petri auditing tool, and persona vectors for tracking model traits. The residency adapts that structure for senior candidates seeking a flexible route into the field.

Four tracks, from oversight to RL systems

• Safety and alignment: Projects may cover scalable oversight, which helps people supervise outputs they cannot efficiently verify; adversarial robustness and AI control, which test whether safeguards hold under attack or attempted evasion; interpretability of model internals; and misalignment model organisms, deliberately constructed models used to study failures under controlled conditions.
• AI welfare: Research may examine whether advanced AI systems could merit moral consideration and how researchers should assess that possibility.
• Capability stress-testing: Residents may map the limits of frontier models in cybersecurity, robotics, cryptography, and other technical domains.
• ML systems and reinforcement learning: Projects may address optimizers, training-data pipelines, compute infrastructure, and reinforcement-learning environment design.

A mentor roster spanning several disciplines

The published mentor roster includes Ethan Perez, lead for Alignment Science; Evan Hubinger, who works on Alignment Stress-Testing; Nicholas Carlini, focused on adversarial ML and security; and Jascha Sohl-Dickstein, a former Google Brain principal scientist working on safety and pretraining. Jack Lindsey, who leads the Model Psychology team, trained in theoretical neuroscience, reflecting the program’s emphasis on expertise developed outside conventional ML career paths.

A trial run for a career switch

Frontier AI labs, which develop some of the most capable general-purpose models, have concentrated much of their recruiting on ML PhDs and experienced AI researchers. The residency widens Anthropic’s pool for safety work and gives the company a structured way to assess how quickly senior specialists can apply their judgment to AI research. Its annualized pay also addresses some of the opportunity cost involved in taking leave from an established role.

For a staff-level engineer, founder, or tenured professor, a six-week placement can test the fit of AI research before a permanent career change. The role remains full time, so even the shortest option may require formal leave or a pause in other work. Anthropic says each resident should finish prepared to pursue full-time AI research at the company or elsewhere.

Applications close November 22

Anthropic says places are limited and directs candidates to the US posting or UK posting. Applicants may also be considered for the Anthropic Fellows Program when that program better matches their experience or proposed work.