Right now, AI models can actually boot up and beat Pokemon autonomously. Yeah, it totally sounds like a really entertaining, harmless party trick. But then you read today's MIT Future Tech Research Report. Right. Those experts warn this exact level of autonomy brings severe dangers. They see a 10 % chance of triggering global societal catastrophe. I mean, we're talking about $100 billion damages by 2030. Beat! Welcome to today's deep dive. I am incredibly glad you're joining
us. We truly have a lot of Complex, fascinating ground to cover today. So we are unpacking an avalanche of massive AI capability jumps right now. Anthropic just dropped their highly anticipated Claude Fable 5 model today. And we really need to examine the structural collision course this creates. Yeah, these blistering capabilities are hitting global security and hardware supply chains. They're also crashing right into our
necessary safety guardrails. So let's start with the sheer raw power hitting the market right now. The newest models arriving today are just absolutely wild. Anthropic released Claude Fable 5 directly to the public recently. It is the first major model from their new mythos family. Fable 5 is absolutely crushing the benchmark charts across the board. It shows truly exceptional results in complex knowledge and vision tasks. It is kind of like stacking Lego blocks of data.
That is a perfect way to visualize its multimodal architecture. You just snap a visual screenshot onto a basic text prompt. And the AI just builds the entire functional code right from there. It does not see images and text as separate languages. It blends them together into a unified understanding of the problem. Which perfectly explains how it can rebuild entire web apps from scratch. Yeah, and it also explains the Pokemon Fire Red achievement. The AI completed the entire game
without any human intervention at all. Just think about this. deep long term planning required for that. You have to manage inventory and navigate highly complex spatial maps. The agent handles all of those sequential decisions completely autonomously. It proves these agents can handle long term enterprise software workflows now. But this incredible autonomous capability definitely comes with a steep price. It costs $10 per million
input tokens right now. Tokens are just the fundamental puzzle pieces of language that AI processes. And output tokens will actually cost you $50 per million. That is exactly double the cost of their previous Opus model. Opus 4 .8 was already considered a fairly pricey option. Startup founders building on Fable 5 will burn cash much faster. They really have to route simple tasks to much cheaper models. Fable 5 is currently available
via their developer API system. API is a software bridge letting different computer programs talk. If you maintain a Pro, Max, or Team subscription, though, you actually get it included completely free until June 22nd. That gives power users a fantastic window to test the capabilities. And this rapid capability leap is mirrored across the whole ecosystem. We are fully entering the agentic era of artificial intelligence. Look at this fascinating new tool called KimiWork,
for example. It deploys up to 300 local AI agents simultaneously. Local agents are AI programs running directly on your personal hardware. They automate complex web tasks and analyze dense financial data sets. Then we have another tool called Macatron, making major waves today. It controls native applications using very simple conversational English commands. We're also seeing Google aggressively push Gemini 3 .5. Their live translate feature handles real -time conversational
translation. Almost perfectly now. You can translate complex, nuanced conversations exactly as they happen globally. The business scale behind these tools is getting incredibly hard to fathom. A company called Lovable just crossed $500 million in annualized revenue. They have seen over 50 million user projects created already. And their usage is accelerating to 1 million new projects weekly. Whoa, imagine scaling to a billion queries. It completely breaks my brain just trying to
comprehend that volume. It is a totally un - unprecedented level of software creation speed. You no longer need to hire an entire team of software engineers. You just describe what you want in simple, plain English. And Lovable spins up the entire working application for you instantly. Two sec silence. It is genuinely magical to watch. With Fable 5 dominating, where does that leave Claude Opus for a user trying to get work done? We actually heard directly from Tariq on the
Claude Code team about this. Oh, wow. What did he specifically say about Opus? Well, he says Opus remains the absolute strongest model for extended work. So it is better for those really long, complex tasks. Right. Opus still handles multi -day, sustained, autonomous execution much better today. So Opus remains the marathon runner while Fable sprints through complex reasoning. Exactly. You always need to pick the right tool
for the job. Which logically brings us to the physical reality underneath all this software. You cannot magically scale to a million new coding projects weekly. It takes immense hardware, massive funding, and incredible physical infrastructure networks. And we are seeing massive structural shifts in the financial markets. The old FAA acronym is completely dead at this point. Financial analysts are heavily using the new acronym MANGOES instead today. That stands for Meta, Anthropic,
NVIDIA, Google, OpenAI, and SpaceX. It perfectly captures the massive financial shift toward artificial intelligence dominance. And the sheer volume of money flowing into these companies is staggering. A startup called Moonshot AI is currently raising $2 billion. That represents a massive 50 % jump in their overall valuation. They were valued at $20 billion just one short month. It is a level of rapid capital accumulation we rarely ever see. But all that investment money has to
buy actual physical hardware. They need these highly specific AI processing chips for 2028. And there is a massive shift happening in global chip manufacturing. Google and Nvidia are reportedly turning to Intel foundries right now. They are actively using Intel as a critical backup to TSMC. TSMC has historically dominated the entire global semiconductor chip manufacturing market. Why are Google and Nvidia suddenly looking at Intel for their 2028 chips instead of just sticking
with TSMC? It largely comes down to intense geopolitical realities and sheer volume. Meaning they just cannot rely on one single factory anymore. Exactly. Relying entirely on a single foundry in Taiwan is incredibly risky. Right. Diversifying away from TSMC is basically a massive supply chain insurance policy. That is exactly what it is for these massive tech giants. Building a semiconductor foundry. It takes many billions of dollars in years. You absolutely cannot just spin up a new
chip factory overnight. It requires massive clean rooms and incredibly complex photolithography machines today. But Taiwan sits in an incredibly precarious geopolitical hotspot right now. If those supply chains get disrupted, the entire AI industry halts completely. Google and Nvidia absolutely cannot afford that massive level of existential risk. So they are pouring massive resources into Intel's new domestic foundries. But the infrastructure... friction is not just
about the physical semiconductor chips. We also have to deeply examine the massive software security vulnerabilities. Microsoft just had to surgically remove over 70 compromised GitHub repositories. Hackers actively slipped highly dangerous malware into these specific coding projects. These compromised projects were tied directly to automated AI coding tools. They specifically targeted automated agent
platforms like Cloud Code and Gemini CLI. It shows how deeply vulnerable these automated development environments truly are. When AI agents write code, they often pull from open source libraries. If a hacker poisons that library, the AI blindly pulls the malware. And human developers are trusting these AI tools a bit too blindly. It is a massive structural security flaw in the modern coding ecosystem. This severe friction extends to the
actual raw coding capabilities as well. A company called Cognition just released their new frontier code benchmark system. The absolute top frontier models currently score just 13 out of 100. That explicitly shows real -world enterprise coding still hits massive walls of friction. I still wrestle with prompt drift myself. Yeah, prompt drift is when an AI gradually loses focus on your original instructions. You ask the agent to fix a specific tiny front -end button issue.
And suddenly it tries to rewrite your entire back -end database architecture instead. We are still very far from having completely autonomous software engineers. The agentic era is clearly here, but it is deeply. deeply messy. We have brilliant flashes of magic mixed with fundamental structural failures. It creates a very precarious foundation for our critical digital infrastructure today. And the mounting geopolitical tensions over chips just adds significantly more pressure.
We are moving incredibly fast on deeply fragile underlying foundation systems. Which brings us to the biggest structural problem we collectively face. The catastrophic risks of this technology scaling without proper safety guardrails. Be sponsored. Before the break, we analyzed intense geopolitical hardware chip supply tensions. And we looked at those massive automated GitHub malware security breaches. These structural software issues lead directly into the broader macro risk
landscape. We have blistering capability growth actively colliding with terrible structural dangers. MIT FutureTech just released a deeply unsettling new AI blueprint today. They comprehensively surveyed 272 leading international AI experts. The detailed findings are honestly enough to keep you awake tonight. They carefully analyzed 24 major AI risks currently threatening our modern society. Experts critically believe 18 of those
risks have catastrophic outcome potential. They assign a greater than 10 % chance to those catastrophic events. That explicitly means... hundred billion dollar damages or massive global societal disruption. And they predict these massive structural failures could easily happen by 2030. That is only a few short, highly volatile years away from us today. The top catastrophic risks they identified are
incredibly specific and deeply concerning. They deeply worry about dangerous autonomous capabilities emerging in these frontier models. They highlight severe geopolitical competitive dynamics aggressively accelerating the global arms race. And they specifically warn about highly weaponized cyber attack capabilities rapidly expanding globally. Which links perfectly back to that recent Microsoft GitHub malware hack. The software vulnerabilities are already
being actively exploited in the wild today. But the MIT study also revealed a massive, terrifying societal vulnerability gap. The people most deeply vulnerable to these catastrophic failures are everyday users. The general public absorbs the massive financial damage when vital... systems break. But the general purpose AI developers are the ones structurally responsible today. Government regulators also hold massive inherent responsibility for preventing these specific
catastrophes. The critical crosshairs are currently aimed at information, finance and national security. Think about autonomous AI getting deeply embedded in critical financial decision making systems. If these AI tools get embedded in the financial sector and fail, who is actually left holding the bag? Well, the MIT study clearly shows the general public takes the massive hit. Wait, really? The public takes the hit? Yeah. While developers and regulators are responsible, they lack financial
incentives to slow down. So every day people suffer. But tech developers and government regulators hold the actual responsibility. That is the tragic, undeniable reality of this massive, glaring accountability gap. The underlying structural dilemma here is incredibly difficult to solve quickly. Any tech company that deliberately slows down for safety takes a massive hit. They immediately lose massive competitive ground in a violently red hot market.
Absent strict external government rules, slowing down is essentially financial suicide today. Developers have deep structural and financial reasons not to prioritize vital risk mitigation. This perfectly explains Anthropic's highly specific approach to the recent Fable release. Ampropic purposefully put a massive safety guardrail system on Claude Fable 5. They have loudly warned everyone about severe recursive self -improvement dangers very recently. An AI system continuously making
itself smarter without human help. So if you prompt Fable 5 with highly sensitive or dangerous topics. The model will actually actively hard block your specific request right away. It automatically routes your session straight back to the safer Opus model. To legally use Fable 5, you also face a mandatory data retention policy. You must formally agree to a strict 30 -day data retention window. Anthropic explicitly promises not to train future models on that retained user data.
But they absolutely need the logs to actively monitor for dangerous emergent behavior. Meanwhile, the completely unrestricted, deeply powerful version is called Claude Mythos 5. But Anthropic intentionally locked that specific model away behind heavily closed doors. It is currently only available for highly approved, heavily vetted security partners now. They are actively taking a massive, deliberate financial hit to ensure global safety. They essentially built the fastest
car but permanently locked the top gear. Because they clearly know the regulatory brakes on the broader industry are failing. It is a fascinating, rare, real -world example of prioritizing structural safety first. But they are largely acting completely alone in this highly specific, cautious approach. Most corporate competitors are just blindly racing toward maximum capability and profit. And Fropic has always actively branded themselves as the
premier safety -first AI lab. They famously split off from OpenAI years ago over fundamental core safety disagreements. So releasing a highly capable autonomous model like Fable 5 is deeply risky. They truly had to thread a nearly impossible corporate and ethical needle here. The mandatory 30 -day enterprise data retention policy is a truly perfect example. Most massive enterprise users absolutely hate corporate data retention
policies today. They desperately want complete privacy and zero data logging for internal... corporate security. But Anthropic stubbornly insists on keeping those specific user logs for 30 days. They need to actively monitor for extremely dangerous emergent capabilities developing online. They are specifically looking for terrifying signs of biological weapons research queries happening. Or highly automated, deeply destructive cyber warfare scripts being actively generated
by their models. Think about the sheer terrifying scale of $100 billion in damages. That isn't just a tiny, easily ignored blip in the global financial market. That represents massive global job losses and entirely decimated corporate pension funds. And a 10 % chance is practically playing Russian roulette with the economy. You would never voluntarily board an airplane with a 10 % crash rate. But we are currently strapping our entire global economy to this exact rocket.
And the ambitious people building the massive rocket currently have no real brakes. Two secs silence. If we logically pull back and look at the broader big picture today. We are living in a deeply profound jarring split screen reality right now. On one side we have. deeply magical, fully autonomous software coding agents. They are completely rebuilding complex web applications entirely from visual scratch today. We have a violently red -hot mangoes economy that is basically
printing infinite money. But on the other darker side, we face a deeply terrifying structural gap. We have a massive, currently unaddressed structural safety and global accountability gap. Actively slowing down to implement crucial safety guardrails takes immense corporate courage today. Anthropic, courageously restricted Mythos 5, but willingly took a massive financial hit. This dangerous competitive dynamic leaves our most critical financial and national security sectors
vulnerable. Completely vulnerable to massive, highly catastrophic global risks by the year 2030. The fundamental corporate incentive structures are completely misaligned with long -term human survival today. Which ultimately leaves you with a deeply unsettling, highly provocative question to ponder. Taking a severe competitive financial hit is currently the only reward for safety. And external federal government regulations are
historically moving far too slowly today. Will it necessarily take the first $100 billion global financial market crash? Or perhaps a massive, deeply weaponized global political disinformation event soon? To finally force the entire global AI industry's brake pedal straight to the floor. It is a truly terrifying, sobering thought to leave hanging in the air. Thank you so much for taking this fascinating deep dive with us today.
We always deeply appreciate you thoughtfully exploring these complex technical topics with us. Stay intensely curious, and we will definitely see you on the next one.