Policy Research, Strategic Advisory
New
We sent a six-digit code to . Enter it below — or use the link in the same email.
Or click the link in the same email — either works.
Enter the email address on your account and we'll send you a link to set a new password.
Remembered it? Sign In
Looking for an AI policy & safety job in San Francisco (SF)? Choose from 25 open roles at 11 employers. Explore real salary data and skills breakdown across remote, hybrid and on-site roles.
Free. No credit card needed.
As of 23 September, 21 AI policy & safety roles are open in San Francisco at 8 employers. 2 new roles have been posted since 21 September. Of the roles that say how they work, 33% are remote or hybrid.
Figures are measured every Monday.
The roles split broadly into technical research and policy design, though many sit at the boundary of both. On the research side, positions such as Researcher, Alignment CoT Monitorability involve running empirical studies on whether chain-of-thought reasoning in frontier models can be reliably monitored, building evaluations, and testing how training interventions affect oversight. Researcher, Alignment Interpretability focuses on mechanistic techniques for understanding model internals, with an expectation to publish findings.
On the policy side, Policy Design Manager, Conventional Weapons requires translating domain expertise into behavioural guardrails, threat models, and escalation procedures. Model Policy Manager does similarly for cybersecurity contexts. More operationally, Research Manager, Biological Safety leads a team building classifiers and datasets to govern biological knowledge in models, while AI Red Team Engineer probes AI coding agent monitors for attack surfaces and builds automated red-teaming pipelines.
Evaluation is the most widely mentioned skill across these postings, appearing in more than half of them, suggesting it is a common thread regardless of whether a role is technical or policy-oriented. Python and agents each appear in roughly a third of postings. A PhD is referenced in only a small minority, and deep-learning frameworks such as PyTorch and TensorFlow appear rarely. Given the small sample, these counts indicate emphasis rather than universal requirements.
The employers' own advertised ranges, for individual roles at different levels — not an average and not a market rate.
Most postings in this set are listed as onsite, making fully in-person work the dominant arrangement stated by employers here. A smaller share are listed as hybrid, with only one posting explicitly offering remote work. No postings have an unstated work mode. Candidates who need flexibility should check individual listings carefully, as the majority of roles appear to expect regular on-site presence in San Francisco. Hybrid terms, where offered, are not further described in the fact pack.
Jobs checked 1 hour ago.
Search by role, company, or anything a posting mentions.
