Anthropic's Safeguards organization builds the policies, evaluations, and detection and enforcement systems that define and hold the limits on how Claude can be used. In this role, you'll lead our policy design team, managing the teams responsible for radicalization, child safety, user well-being, harmful manipulation, and election integrity, among other harm areas.
The team is responsible for understanding and defining the risks that come with engaging with Claude, how those risks materialize in the real world, and the mitigations needed to prevent them. As the manager, you'll work with your team to draw the boundaries between what is and is not allowed, then partner with research, product, and engineering to build the right interventions. Mitigating these harms takes the whole stack: the values and judgment trained into the model itself, the policies and detection systems we enforce on top of it, and the interventions we build into our products. More capable models, new product surfaces, and new user behaviors will keep testing these boundaries, so the team's policies have to keep pace.
You'll work closely with product to develop and iterate on the strategy and vision for how our safety layers fit together, and with your team and cross-functional partners to decide which mitigations make the most sense and how to implement them. You'll also coordinate policy decisions across the portfolio: ensuring they're made with the right stakeholders in the room, tracked over time, and applied consistently across harm areas and product surfaces.
This is a leadership role for someone who combines expertise in the harm areas themselves with fluency in how frontier models are actually developed and deployed, and who does their best work across team boundaries. You'll spend as much time developing the leads who own each harm area as you will on the policy questions themselves.
*Important context for this role: some of the work involves exposure to explicit content, including material of a sexual, violent, or psychologically disturbing nature.
Key responsibilities
• Lead, develop, and grow the managers and teams responsible for the consumer harms portfolio, including child safety, user well-being, harmful manipulation, and election integrity
• Coordinate policy decisions across the portfolio, and build the mechanisms that keep them tracked, consistent, and legible — so stakeholders know what was decided, why, and who owns what
• Set the strategy for how mitigations built on top of the model — policies, detection and enforcement systems, and product interventions — complement what is trained into the model itself, partnering closely with the alignment training team that owns Claude's character
• Prioritize across harm areas competing for the same resources, and make those tradeoffs and their rationale clear to leadership
• Serve as the escalation point for high-severity and ambiguous consumer harms decisions, including rapid response to emerging risks
• Partner with engineering, data science, product, legal, and research across the model development cycle so consumer harms considerations are represented from training through launch, on every surface where Claude is deployed
• Engage external experts, civil society organizations, and regulators, and translate that engagement into stronger policy and enforcement
Minimum qualifications
• Experience leading teams — including managing managers or senior specialists — in AI safety, product policy, or a related field
• Deep, applied familiarity with consumer harm areas such as child safety, mental health and well-being, manipulation, or election integrity, and good judgment about how these harms differ in mechanism, severity, and mitigation
• A track record of exceptional cross-team collaboration: building durable working relationships with teams you don't control, and getting to shared decisions where ownership is genuinely distributed
• Working understanding of how frontier models are developed and deployed — the training and fine-tuning cycle, evaluations, and launch processes — and how different model environments (consumer products, APIs, agentic tools) change both risk and the mitigations available
• Experience translating policy positions into mechanisms that can be enforced and measured, and communicating the reasoning to technical and non-technical audiences, including executives
• Sound judgment in ambiguous, high-consequence decisions, and comfort making a call and escalating appropriately on incomplete information
