Google DeepMind's AGI Safety team is hiring for existential risk roles

Google DeepMind's AGI Safety and Alignment Team (ASAT) is actively recruiting for multiple roles aimed at reducing existential risks from advanced AI systems. Led by Rohin Shah, the team operates with a distinct focus on preempting severe harms from highly capable future AI systems, rather than just current deployments. Their work spans alignment research, governance-enabling technical development, and stress-testing advanced AI capabilities to identify latent risks.

The team differentiates itself from other safety groups at competitors like Anthropic and OpenAI by emphasizing technical enablers for governance, preferring evidence-based knowledge over advocacy, and prioritizing research geared toward much more capable future systems. ASAT has pioneered efforts like the first suite of dangerous capability evaluations and methods to measure AI progress over years, such as the Epoch Capability Index. They aim for coherence in their overarching approach, publishing technical roadmaps like 'An Approach to Technical AGI Safety and Security' to guide their multi-bet research strategy.