Google DeepMind's ASAT team hires for AGI safety research in SF and London

Google DeepMind's AGI Safety and Alignment team (ASAT), led by Rohin Shah, has opened hiring for multiple technical staff roles as of July 2026. The call, posted on LessWrong and the AI Alignment Forum by researchers Sebastian Farquhar, Rohin Shah, and Neel Nanda, spans all research areas covered in their recent work. ASAT focuses on risks from more advanced AI systems, including aligning AGI, defending against misaligned deployments, and supporting coordinated safety. While locations are flexible, most roles will sit in San Francisco or London. Newly emphasized domains include deep alignment and stress testing, created in response to increased capabilities.

ASAT differentiates itself from safety teams at Anthropic and OpenAI by prioritizing technical enablers for governance, preferring knowledge over advocacy, and focusing on future superhuman systems rather than current models. Concrete examples include the first suite of dangerous capability evaluations, a metric for identifying latent reasoning architectures, and a method powering the Epoch Capability Index. They also conduct conceptual research like MONA, debate, and honeypot evaluations. Their aim is to produce evidence that resolves factual disagreements and improve execution for alignment, feeding into unified roadmaps like the GDM AI Control Roadmap. This hiring push signals a strategic expansion of preemptive safety research at one of the world's leading AI labs.