Keeping AI in a Box

YouTube · Watch

Robert Miles expresses regret over his past framing of the “AI in a box” debate, arguing that the real reason it won’t work isn’t just theoretical escapability by a superintelligence, but humanity’s practical failure to even attempt containment.
The “AI in a box” concept refers to sandboxing a superintelligence to limit its actions, a debate where his past argument was that such an AI would inevitably “figure out a way to get out of the box” by exploiting human inability to “reliably differentiate between outputs you want to allow and ones you don’t.”
The more immediate and pressing danger is that developers are “connecting these things up to everything as quickly as possible with basically no safeguards,” rendering the theoretical debate about AI escape moot by not even attempting to box it.

A cluster of shorter, lower-production advocacy and link-sharing videos in the same window, reflecting an increasingly activist posture rather than pure technical explainer content: People are Catching On (open letter + NewsNation collaboration), Don’t Let “Scepticism” Make You Useless: ActOnAI.org, Did ChatGPT Just Commit a Felony? (QuitGPT collaboration, calls to contact elected representatives), Is China Open to Making a Deal on AI?, “Stop the AI Race” March in SF and protest outside OpenAI (both 2026-07-11), Mutually Assured Compute Destruction (AI 2040: Plan A framing), plus resource-pointer videos Useful Websites: AISafety.com, AI can’t DO my JOB quite yet, AI Lie Detectors?, and Why I Avoid Sci-Fi Specifics.

Emerging pattern: Compared to prior gathers, this window shows a visible shift toward direct advocacy (street protests, calls to contact representatives, named campaign collaborations like QuitGPT and ActOnAI.org) alongside the usual technical-explainer content — worth watching whether this is a durable posture change or a temporary response to a specific news cycle.