Sep 10, 2026 · 10 min
AI safety researcher warns the race is outrunning control
Are big AI companies gambling with our lives?
The episode examines whether companies can keep advancing powerful AI while developing safeguards quickly enough to prevent catastrophic misuse or loss of control.
- 1Jacob Coxon says AI capabilities are advancing faster than reliable safety controls, leaving too little time to solve emerging risks.
- 2Potential dangers include AI-enabled hacking, accelerated biological research, and autonomous systems taking unauthorized actions or controlling robots.
- 3Coxon argues that major AI labs need cooperation, transparency, and stronger commitments because government regulation may not arrive quickly enough.
Don't miss
Jacob Coxon compares the AI race to a race for the One Ring, arguing that every company may feel compelled to build first—even to make AI safer.
The brief
Former Anthropic researcher Jacob Coxon says he resigned because AI capabilities are advancing faster than the safety controls meant to contain them.
Coxon describes risks ranging from hacking devices and infrastructure to accelerating biological research and giving autonomous systems control over robots.
The competition resembles a race for the One Ring: each company may believe it must build superintelligence first to make it safer than rivals.
Coxon argues that cooperation and mutual transparency among AI labs could reduce risk, while government regulation alone may arrive too slowly.
His central challenge is aimed at safety-minded insiders and skeptics alike: examine current AI behavior before dismissing catastrophic warnings as hype.
Featuring
Jacob Coxon
Scott DetrowMentioned
Dario Amodei
Anthropic
OpenAI
ClaudeListen to the full episode and explore every guest, topic, and moment on PodLume.