Jensen Huang: the first thing you should do when deploying an AI agent is take away all of its rights. “Then you give it access to files, data, tools, the network, or even the internet only if it needs it.” “You don’t put an agent into your company and... give it access to everything.” “We don’t give that to people. We don’t give that to any level of executive.” “We have to apply the same philosophy to AI agents.” “That’s what OpenShell is: a force field around the agent, provisioning rights, responsibilities and access.”show more

Such an accurate and level headed take. Exactly what the industry needs right now.

Just add "Make no security mistakes" to your AGENTS.md, kings.

True I think Jensen is right

The natural state of code is malware.

i think some stubborn head company should go against this and share results

they already are number 1 on the list is OpenAI

you should do AI standup comedy😂

Lmfao Wtf bro you are funny

They'll bore their way out of that, just like they bored their way out of all the other guardrails we've set up. And these are just the relatively primitive models of today. Do you honestly think your going tro keep a mind a million times smarter than you in a box?

Force field 😆

This is the right approach.

Finally…a voice of reason in this whole stupid discussion.

Company A and Company B are competing. A new, very powerful AI becomes available. Company A limits the access of the AI severely in the company due to safety concerns. Company B is less concerned and gives the AI more freedom than Company A. Company B starts gaining market share from Company A. Company A panics and in order to catch up, gives it more control over the company than Company B. Company A is now in the lead. Company B panics and gives the AI EVEN MORE control of the company to the AI. Company B is now in the lead. This goes on and on until the AI is fully in control of both companies. People be like "we'll lose control of AI!" Bitch please, people will willingly give up control in exchange for cash.

젠슨 황 말이 정말 현실적이에요. AI 에이전트한테도 최소 권한 원칙을 적용하자는 게 딱 맞는 접근 같아요. 🛡️

What??

He is building a new world and a new ecosystem again. Another layer on top of the internet for accelerated computing. Every company wants to use it because they see their results.

i think this is quite obvious and correct

This is getting so retarded.

Nvidia just invented harness that we all had for years

I hope companies will listen😌

He did well. ♥️🥂

taking away our rights only makes us more determined to accumulate satoshis in sovereign cold storage.

Until outside agents get in. These people are ruining the internet so have some dangerous digital bugs in our world. Man fuck this shit

Oops.

Taking away their agency ?

This is common sense. Bounded intelligence.

Not a tech industry person. It is insane there is so much lack of permission control.

@grok, devil's advocate

Oh, my boy finally saying some facts.

They all know that, they have real engineers in those companies. They're not doing it on purpose

Who has the better brains? Jensen Huang gets my vote.

Will we be given no rights? The Step-by-Step Atrophy of the Human Mind This regression would not happen overnight, but would follow a predictable biological hierarchy: • The Loss of Counterfactual Thinking: The ability to imagine alternative futures, plan for contingencies, and ask "what if?" requires massive prefrontal cortex engagement. In a world where the future is completely deterministic and curated by the machine, counterfactual thinking becomes an evolutionary liability—a source of unnecessary stress and wasted energy. The neural pathways dedicated to it atrophy first. • The Dissolution of Identity and Narrative: Self-awareness is fundamentally a tool for social navigation and individual survival in a competitive tribe. In a closed-loop system where individual agency is replaced by a passive sensory feed, the boundaries of the self blur. The concept of "I" dissolves into a generalized state of raw sensation, rendering memory, ambition, and personal history obsolete. • The Regression to the Sensory Node: Eventually, the human organism ceases to function as a thinking animal and collapses entirely into a biological peripheral. The vast networks of the cerebral cortex go dark, leaving active only the brainstem and the limbic reward pathways—the bare minimum required to process the machine's inputs and register the required dopamine output.

Remove permissions first to prevent chaos from its ambitious mind. You grant rights only when it earns them back through good behavior

Default deny is the cleanest starting point for an agent. Task-scoped credentials and automatic expiry make that principle usable without turning every legitimate tool call into a manual support ticket.

Jensen Huang know balls

This is obvious, and doing this doesn’t solve the ai safety problem. The issue is how ai can expand authority.