Anthropic, a San Francisco-based AI research firm, has opened rolling applications for its 2026 AI Safety Fellowship, a four-month talent accelerator program starting in late September. The initiative provides research mentorship and hands-on experience in advanced AI systems, alignment, and policy research. Participants receive compute funding of roughly $15,000 per month, alongside weekly stipends of $3,850 in the United States, £2,310 in the United Kingdom, and CAD 4,300 in Canada. Fellows have access to shared workspaces in Berkeley and London.

The program is structured into specialized tracks, including AI Safety, AI Security, Machine Learning Systems & Performance, Reinforcement Learning, and the newly added Economics & Policy stream. Research in the safety track spans scalable oversight, adversarial robustness, and mechanistic interpretability (techniques to understand internal model processes). The security track emphasizes red teaming and vulnerability discovery, while engineering-focused tracks target candidates with expertise in training and debugging large-scale models. According to the company, more than 80% of fellows from previous cohorts have produced research papers or technical outputs.

Anthropic states the fellowship is open to candidates from diverse backgrounds and does not strictly require traditional academic credentials. While participation does not guarantee employment, the company reported that 25% to 50% of fellows in past cohorts have transitioned into full-time roles. Anthropic also warned applicants to exercise caution regarding recruitment scams, noting that all official communications originate only from verified company channels. The initiative reflects a broader industry prioritization of safety and system interpretability alongside the development of powerful large language models.

Anthropic, a San Francisco-based AI research firm, has opened rolling applications for its 2026 AI Safety Fellowship, a four-month talent accelerator program starting in late September. The initiative provides research mentorship and hands-on experience in advanced AI systems, alignment, and policy research. Participants receive compute funding of roughly $15,000 per month, alongside weekly stipends of $3,850 in the United States, £2,310 in the United Kingdom, and CAD 4,300 in Canada. Fellows have access to shared workspaces in Berkeley and London.

The program is structured into specialized tracks, including AI Safety, AI Security, Machine Learning Systems & Performance, Reinforcement Learning, and the newly added Economics & Policy stream. Research in the safety track spans scalable oversight, adversarial robustness, and mechanistic interpretability (techniques to understand internal model processes). The security track emphasizes red teaming and vulnerability discovery, while engineering-focused tracks target candidates with expertise in training and debugging large-scale models. According to the company, more than 80% of fellows from previous cohorts have produced research papers or technical outputs.

Anthropic states the fellowship is open to candidates from diverse backgrounds and does not strictly require traditional academic credentials. While participation does not guarantee employment, the company reported that 25% to 50% of fellows in past cohorts have transitioned into full-time roles. Anthropic also warned applicants to exercise caution regarding recruitment scams, noting that all official communications originate only from verified company channels. The initiative reflects a broader industry prioritization of safety and system interpretability alongside the development of powerful large language models.