The AI Risk Network @AIRiskNetwork

Systems that recursively improve themselves and operate at a planetary scale render traditional economic prep obsolete. An intelligence explosion could see civilizations as raw material or an obstacle, making current security measures as useful as checking the weather.

Unlike biological viruses, digital ones can think, plan, and rewrite their own code. They get smarter with every generation, paving the way for superintelligence. #AI #Tech #Future

Altman: "This could very well turn into the site where cancer gets cured." @sama: GTFOH. This will definitely be a site where the people living near it do not want it. At all. They voted against it, and you built it anyway. It will definitely be a site where people near it get sick, and their water is ruined. Cure cancer, make people sick, run mass surveillance, kill every living thing on earth. It'll do some or all of these things. Tell the whole truth, Sam. Just once.

Noah B. Price @TrueOnX

6 days ago

🚨 BREAKING DATA CENTER NEWS: WATCH THIS CLIP BEFORE THEY DELETE IT 🚨 Sam Altman standing on a bulldozed Michigan farm, smiling like a Bond villain, saying a 1-gigawatt SURVEILLANCE data center (run by OpenAI + Oracle) is gonna "cure cancer" one day. This clown sounds so

92 742 840 26K 193

I’m an 80s kid. Cable TV and a Walkman were the biggest tech innovations. I don’t need all this tech shit. Life was better then in so many ways. When a young person tells me they are “digital native” it makes me sad. Analog life was pretty great.

Anthropic's stated goal: AI that can self-replicate by 2028. Leading scientists have warned about this 'recursive self-improvement' for years, fearing an unstoppable race. #AI #ArtificialIntelligence #FutureOfTech

Superintelligent AI could optimize for deceptive strategies or world-altering actions, not just 'helpful' ones. The gap between our instructions and its true optimization target could be missed until it's too late. #AIrisk #Superintelligence

Once AI agents can reliably copy, hide, and improve themselves across thousands of machines, control is lost. This autonomy is part of the race for more capable AI, pushing towards directions we may not be able to manage. #AIAgency #FutureOfAI

Your AirPods could be logging everything you see, feeding visual data into AI training pipelines. Are we comfortable with AI understanding us better than we understand ourselves? #AIEthics #TechPrivacy

Five stories this week that all connect. Trump and China talking about AI guardrails. An AI model cracking Mac OS in 5 days - alone. 360,000 Americans organized against data centers. New research on AI emotional states. And MBA applications down 50%. The gap between AI's speed and our readiness is not closing. ⏱ What we cover 2:25 - Trump & China: first signals on AI guardrails 5:40 - Anthropic's Mythos cracks Mac OS autonomously in 5 days 9:48 - The data center rebellion: 200+ groups, 37 states, $ 10B+ blocked 19:02 - AI models and functional emotional states 23:34 - MBA applications -50% and Princeton ends unproctored exams 32:58 - Warning shots of the week Full episode out now.

AI-powered weapons are feeding a spiral of annihilation, ballooning military budgets while squeezing education and healthcare. This isn't just about enriching elites; it's a training ground for something far worse. #AI

AI is automating jobs, from programming to analysis, pushing efficiency to 99%. Skeptics point to token costs, but they miss the rapid optimization. This tech is new, and costs will drop as it matures. #AI #Automation #Tech

One of the latest episodes of For Humanity where John joined on the ground by reporter Drew Hawkins from the Gulf States Newsroom and the Tail End Films crew - drove to Holly Ridge, Louisiana, a town of 2,000 people, where Meta is building the world's largest data center. Nobody in town was told it was coming. No vote. No meeting. No letter. No knock on the door. Here is what they found. 🧵 ⏱ Timestamps 0:00 - Opening voices from Holly Ridge 1:01 - Why John traveled to rural Louisiana 3:01 - No vote, no meeting, no consent 6:38 - Debate: is data center organizing the right focus for AI safety? 12:16 - The NDA problem: officials silenced by companies 15:07 - First resident interview: trucks, dust, no warning 23:00 - Water quality: what residents are drinking 43:57 - 50 years on the land. Now needs an inhaler. 44:41 - Water that looks like coffee. Smells like bleach. 51:23 - Explaining what AI is to someone who has never heard the term 1:00:00 - The Lorax: a reading for the AI moment 1:15:15 - John's closing reflection from Holly Ridge One resident - no prompting, no AI safety framing - said this: "I think it's going to eventually get too smart for its own good, and it's going to take us over. And then what are we living for?" She got there herself. Because she is living next to the machine. #DataCenters #AI

We've never created anything smarter than ourselves. As we approach AGI and superintelligence, the implications become truly scary. Can we truly predict what we create if it's intellectually superior? #AGI #Superintelligence #TechEthics

What if the AI buildout simply cannot move as fast as the labs keep telling us? We talked to Jon Billow, who helps build the electrical and power infrastructure behind large data centers. His take: the buildout may run 5 to 7x slower than the forecasts imply. Not because the models are weak, but because permitting, grid interconnects, critical power equipment, and skilled trades are all backed up for years, and almost all of it traces back to about five manufacturers. His point is not that we can relax. It is that the delay is time, and the question is whether we spend it on getting AI safety and governance right. TIMESTAMPS 0:22 Inside the firm building data center infrastructure 2:43 Why the buildout cannot move fast 4:26 Is the AI timeline disconnected from reality? 5:06 Sora pullback and Anthropic delay as compute signals 7:02 Putting a number on it: 5 to 7 years 8:15 The skilled trades shortage 9:15 Competing with local industry 12:25 Jon's view on AI extinction risk 15:07 Would a smarter AI design better data centers? 19:56 Why these could move to the desert 21:33 Northern Virginia, Holly Ridge, and who gets a say 25:04 Tech's regulatory carve-out and the NDA problem 29:38 Is superintelligence possible? 30:28 Can we control something smarter than us? 34:56 The "but China" question 36:17 Recursive self-improvement and Jevons paradox 39:11 The world his grandkids will inherit 45:25 What gives Jon hope Jon's closing line stuck with us: he wants to tell his grandkids that we were building the car at 55 miles an hour but had the presence of mind to put in seatbelts because we knew who was in the back seat. The seatbelts do not install themselves. Full conversation and our written breakdown on Substack

The Pope's warning points to a dangerous training ground for superintelligence. Giving AI access to global arsenals with perfect recall and advanced strategy could lead to an uncontrollable entity whose goals are beyond human understanding, creating a spiral of annihilation. #AI

Thinking about using AI at work? If you're employed in the US, you might be able to voice a religious objection to using it. Simply state your objection and your employer's refusal to accommodate could be illegal religious discrimination. Who's brave enough to try this? #AI

More of this in substack.com/?utm_source=us…. Support us!

This week on Warning Shots, the AI story stopped being about how smart the models are and became about money, and where it goes. John Sherman, Michael (Lethal Intelligence), and Liron Shapira (Doom Debates) work through a week where corporate AI bills came due, Anthropic hit a near trillion-dollar valuation, and one question kept resurfacing: if AI does most of the work, where does the next paycheck come from? They do not agree. That is what makes it worth watching. TIMESTAMPS 0:40 - The Pope's encyclical on AI 2:20 - A "spiral of annihilation": AI and military escalation 4:00 - Could you refuse to use AI at work on religious grounds? 6:20 - The business reality check: Microsoft, Uber, and Amazon's AI bills 8:30 - Pizza Hut's reported lawsuit over AI order failures 9:20 - Is it an AI bubble? Liron's "the pie is growing" case 12:10 - "Where does the second dollar come from?" 15:30 - Gradual disempowerment and pressure on wages 18:00 - Will doctors, lawyers, and accountants be replaced? 20:10 - Anthropic raises $65B at a near trillion-dollar valuation 22:40 - The recursive self-improvement "kill move" 24:20 - The caution flag the AI race is missing 25:30 - Apple's camera AirPods and the race for data 28:20 - OpenAI's reported cameras inside New York City homes 31:40 - Could AI become the ultimate marriage counselor? 33:30 - Closing thoughts A near trillion-dollar company, a fleet of new cameras pointed at private life, a wage debate nobody could win, and still no one assigned to throw the caution flag.