This conversation with Ziyue Wang explores some of the most pressing questions at the frontier of AI safety: what it means for powerful AI systems to be safe enough to deploy, how companies model misuse risks such as cyberattacks, CBRN threats, manipulation, loss of control, and malicious actors, and why problems like AI sandbagging remain difficult to detect and prevent. Ziyue also reflects on the broader societal impact of AI, the gap between capability development and safety work, the role of regulation in the EU, U.S., and China, and the personal journey of pivoting into frontier AI safety.
Ziyue is an AGI safety researcher and Tech Lead. He works on frontier safety mitigation, with a focus on safeguards against cybersecurity threats, biological risks, and other high-stakes misuse scenarios. Daily he responds to the question: how to make increasingly powerful AI systems safe enough to deploy responsibly? Before moving into AI safety, Ziyue built strong technical foundations through work in quantitative research, software engineering, large language models, and coding. He also writes about AGI safety and the future of AI through his Substack,
Belay the Future: https://ziyuewang.substack.com/
You can also listen to our conversation here:
Digital Kallipolis is a space for conversations about AI, democracy, human dignity, and the future we are building together. If you believe technology should serve people, and that informed voices can shape a safer, more just digital world, please SUBSCRIBE!
Thanks for stopping by! “Remember [Our] Humanity, and forget the rest…” —Russell–Einstein Manifesto, 1955
Digital Kallipolis 