Schmidt Futures - Footnote 57

Note: The source does not mention the grant amount of $500,000. The claim mentions risks like AI systems escaping laboratory control and self-replication, which are not explicitly mentioned in the source. The source only mentions Kang's work on cybersecurity capabilities of AI agents. The source does not explicitly state that Bo Li is developing virtual environments with red teaming algorithms to evaluate appropriate access levels for AI systems, but rather to evaluate AI systems and AI agents, with a focus on exploring what level of access to AI models is needed for different levels of evaluation.