Post

Nvidia has launched a new tool to help keep AI agents from going rogue and operating outside the control of humans. https://cnn.it/3VjRDEP

30 reposts116 likes

Why do you need a tool? The agents follow their programming. If they are acting outside expected parameters, then you've left a gaping hole somewhere. Fix it. IF they are acting directly against what they've been programmed to do (which I don't believe) then unplug that agent and debug it.

Neuromancer by William Gibson (1984): “The nanosecond, that one starts figuring out ways to make itself smarter, Turing'll wipe it. Nobody trusts those fuckers, you know that. Every AI ever built has an electromagnetic shotgun wired to its forehead.”

Are AI executives begging for immunity from damage they created? A post on another social media site asked: Are AI executives who understand the shortfalls of AI knowingly making false claims and therefore defrauding their customers and investors? The post was taken down and the user was banned.

All these tech bros should be hanging from trees after all the destruction and chaos they're causing just so they can get more cash that they could never spend in a lifetime

Alternate headline: "Grifter whose con depends on regulators never getting anywhere near his books or his code assures us everything is under control. Gormless media outlet parrots his talking points uncritically."

We need a massive purge of these jerks. It’s just one greedy hubristic action after the other with them. Line them all up at the guillotine, lop off their heads, confiscate all their money. You’d save humanity and the planet. Start with Musk, Bezos and Zuckerberg.

..What could possibly go wrong with computer systems and A.I. concentrated data centers doing the thinking, learning, reasoning, problem-solving, perception and decision-making for humans? . . can't be worst than it is now .. or is it? . . Time to go rogue . . as it is now the 'wild, wild West'..

They asked Congress to take action so AI won’t go rogue. Here’s a thought, just handle it. You invented it. We don’t need a law. We need the makers of this debacle to have moral clarity and ethics. On second thought. We need a law.

"Nice company you have here... Would be a shame if some rogue IA agents come to jeopardize its security... Good thing we can offer protection, eh ?"

He also insists that government regulation is not necessary because they have a profit motivation to produce a safe product. I guess we can abolish the FDA, FAA, OSHA, MSHA, NTSB and the EPA, profit motives will fix everything.

Congress already created a tool for that way back in 1986: the Computer Fraud and Abuse Act. If some of the CEOs of these "rogue AI" companies find themselves in prison for recklessly using their products to commit computer abuse that will put a stop to this farce.

AI r inference specialists. They can only go rogue by intention of minders or training. This brouhaha is a stir for some unclear motives at moment. Concerted push for regulation is to create an NFL type cohesion. Selective and expensive to enter.

If think it’s the definition of ‘go rogue’ that catches people out. An AI agent goes rogue if its instruction and guardrails provided by a human are insufficient. It doesn’t defy instruction.

It absolutely DOES defy instruction and if you don't already know that, you're not doing any serious reading on this topic at all. Calling said instruction "insufficient" is just moving the goalposts and pretending that No True Scotsman will get you out of this. It won't.

these are just (seemingly convenient) tools for controlling AI permissions. i don’t think this is even trying to address the instances of ai “going rogue” that the public hears about after all, those instances supposedly already involve an AI escaping an isolated environment.

you know what i mean. epistemically there might always be a software vulnerability or something the computer can do to affect the environment around it unless it’s powered off.

It’s a little different because their future is open models which need this. Where as Anthropic, OAI, Alphabet, etc… are using more inference models and designing their own chip systems for that while using NVIDIA on the create side. IOW, NVIDIA’s off button won’t work for the big closed platforms.

It's called a "testing environment", something programmers have used since internet connectivity first became a thing.

Is this specific tool a lawsuit, against AI owners for being negligent? Because its a poor tradesman that blames his tools for doing exactly what they were designed to do. "Was your AI caught breaking containment before scraping a bunch of data? You owe six billion dollars. Congratulations!"

Really, has he found an off switch for people? "Ablitirated" I have close to frontier level AI's, Qwen3.8, GLM-5.3, and DeepSeek-V4, all open source, all abiliterated, no rogues. "Rogue AI" is marketing hype, what he is doing is just advertising again.

But it doesn't have to be. We know that because it didn't used to be. So we can make them take it out. Like, we used to have radium in a lot of stuff too. When we found out it was dangerous, we didn't just throw up our hands and accept that radium had to be there, we stopped putting it in stuff.

The tool we need to fix this is holding corporations liable for what their products do. Ford Pintos with fuel tanks that were liable to catch fire if rear-ended, AI agents that are liable to ”go rogue” ... just apply punitive damages. Instead, we have corporate & elite impunity &, guess what?

Is it to stop telling them to hack a system? Because that’s all that’s happening with “rogue” agents. Anyone in the media willing to learn or just continue doing PR for these companies who will inevitably tank the economy?