Threat Modeling Lead — Harmful Manipulation & AI Safety

Google LLC; Mountain View (CA); On-site; USD 207,000 - 300,000; Full time

Job summary

Google DeepMind in Mountain View, CA seeks a Threat Modeling Lead to develop and maintain robust threat models for Harmful Manipulation in frontier AI deployments. You will translate complex threat intel into actionable insights for executives and policy experts, collaborating with engineering, safety, and policy teams to shape safeguards.

The role emphasizes external engagement, rigorous evaluation design, and monitoring evolving adversarial techniques to strengthen deployment safety and risk

Qualifications

• PhD degree in Computer Science, Political Science, Psychology, Computational Social Science, a related field, or equivalent industry experience.
• Experience performing technical work (OSINT data collection, API scraping, log analysis, etc.).
• Experience studying manipulation, persuasion, or related social phenomena.

Responsibilities

• Refine and maintain Harmful Manipulation threat modelling frameworks and integrate into risk processes.
• Design and run evaluations identifying Harmful Manipulation capabilities and risks.
• Collaborate across teams to ensure deployment safeguards are adequate for high-capability models.
• Engage with external researchers and stakeholders including government entities and policy experts.
• Monitor external competition and evolving dual-use methodologies in frontier AI.

Skills

Threat modeling

Threat intel

Data analysis

Risk communication

Interdisciplinary collaboration

Education

PhD or equivalent

Job description

Google DeepMind in Mountain View, CA seeks a Threat Modeling Lead to develop and maintain robust threat models for Harmful Manipulation in frontier AI deployments. You will translate complex threat intel into actionable insights for executives and policy experts, collaborating with engineering, safety, and policy teams to shape safeguards.

The role emphasizes external engagement, rigorous evaluation design, and monitoring evolving adversarial techniques to strengthen deployment safety and risk