WorkSeptember 9, 2026via TechCrunch AI
ControlAI’s Connor Leahy on why superintelligence is ‘not a weapon, it’s an adversary’
Why it matters
As AI capability outpaces control mechanisms, safety researchers and policymakers are reframing the superintelligence problem from technical containment to adversarial governance. This shapes how enterprises think about AI deployment risk and regulation.
Key signals
- Connor Leahy is U.S. Executive Director of ControlAI
- Recent safety incidents cited: OpenAI's Hugging Face breach
- Frame: superintelligence as adversary vs. weapon
- Core tension: AI systems more capable than human control
- Published Sep 09 2026 (future date — verify publication)
- Episode references OpenAI's Hugging Face breach as a safety incident
- Topic: controllability and reliability of superhuman AI systems
- Framed as superintelligence as an 'adversary' rather than a controllable tool
- Podcast format — opinion/commentary, not breaking news or a specific development
The hook
Connor Leahy: superintelligence isn't a weapon—it's an adversary we can't control yet.
AI companies have been talking about superintelligent AI like it’s inevitable, but recent safety incidents like OpenAI’s Hugging Face breach are demonstrating the potential dangers of deploying AI systems that are more capable than humans. So what happens when we can’t reliably control what these sy…