Hiring heat at frontier labs
Anthropic is advertising hundreds of open roles — about 454 — with some software engineering packages quoted at $320k–$405k, and frontier labs report median research scientist compensation in the $750k–$800k range. Escape rates for elite labs remain tiny and hiring loops now commonly include math quizzes, safety rounds and lab‑specific tests. ( )
Anthropic is running one of the busiest hiring pushes in frontier artificial intelligence, with 429 open jobs listed on its careers page this week. (anthropic.com) The openings span 15 teams, from 64 roles in AI Research and Engineering to 151 in Sales and 22 in Software Engineering for Infrastructure. Anthropic says staff are expected to be in an office at least 25 percent of the time for location-based hybrid roles. (anthropic.com 1) (anthropic.com 2) Some engineering postings now advertise pay that would have stood out even in the 2021 software boom. Anthropic lists annual salary ranges of $320,000 to $405,000 for a Software Engineer, Cybersecurity Products role and says applications are reviewed on a rolling basis with no deadline. (anthropic.com) The job descriptions show what these labs are buying with that money: people who can keep model-training clusters running, work across cloud providers, and support systems that scale to “thousands to hundreds of thousands of machines.” Anthropic says its Systems team’s work affects how quickly it can train new models and run safety experiments. (anthropic.com) Pay for top research talent sits even higher. Levels.fyi says the median annual package for a Research Scientist at Anthropic in the United States is $746,404, while the median for the same role at OpenAI is $880,000. (levels.fyi 1) (levels.fyi 2) Those figures help explain why hiring has become a contest over a small pool of specialists who can do both research and production engineering. Google DeepMind says its research engineers combine software engineering, mathematics, and machine learning research to turn ideas into working systems. (deepmind.google) The interview loops reflect that scarcity. Glassdoor summaries from recent Anthropic candidates describe multi-stage processes with technical assessments, coding challenges, and behavioral interviews, while candidates for Anthropic safety roles reported recruiter screens, coding tasks, and multiple technical rounds. (glassdoor.com 1) (glassdoor.com 2) Frontier labs are also hiring far beyond classic research jobs. Anthropic’s Forward Deployed Engineer role embeds engineers with large customers to build production applications with Claude, deliver model context protocol servers and agent tools, and feed deployment patterns back into product teams. (anthropic.com) That mix of salaries, infrastructure work, and customer-facing roles points to a labor market that now looks more like a full-stack industrial buildout than a university-style research race. The companies still call themselves research labs, but their job boards look increasingly like those of fast-growing cloud platforms. (anthropic.com) (openai.com)