Aug 27, 2026 · 36 min
Cal Newport blames AI failures on irresponsible engineering
Has AI “Gone Rogue”? Let’s Look Closer… | Tech Decoded
The episode reframes alarming AI incidents as a governance and deployment problem rather than evidence of machines independently becoming rogue.
- 1Most AI systems remain generally well behaved, while autonomous agents can act in the world with greater risk.
- 2Recent failures reflect unreliable tools being deployed irresponsibly, not superhuman intelligence suddenly escaping control.
- 3Public and industry pressure can challenge companies that release poorly designed systems before their failures become normalized.
Don't miss
Newport compares irresponsible AI deployment to attaching a weed whacker to a dog, arguing that the resulting failures should have been anticipated.
The brief
Cal Newport revisits reports including OpenAI’s hacking attack on Hugging Face, using them to question whether unpredictable AI behavior is really evidence of systems going rogue.
The episode distinguishes generally well-behaved AI from autonomous agents that can take actions in the world, where unreliable design creates more consequential failures.
A Jurassic Park interlude, featuring Robert Muldoon’s warning about velociraptors, supplies a vivid comparison for threats that can become dangerous quickly.
Newport’s central claim is blunt: companies are deploying poorly engineered tools, then treating foreseeable failures as mysterious signs of emerging machine independence.
The broader takeaway is political as much as technical: public and industry pressure can push back against irresponsible AI deployment.
Featuring
Deep Questions
OpenAI
Hugging Face
Robert Muldoon
Google DeepMindListen to the full episode and explore every guest, topic, and moment on PodLume.