Too dangerous to release 🔒🤖

🛡️ Anthropic Says Mythos Is Too Powerful for You Peasants

Anthropic's newest model Mythos is apparently so good at finding security exploits that they're locking it behind an enterprise-only program called Glasswing - sharing it exclusively with AWS and JPMorgan Chase. OpenAI is reportedly eyeing a similar approach. Critics say this is also a convenient way to block competitors from distilling their models while printing enterprise contract money.
Why it matters: The era of "here's our best model, go nuts" might be over. Whether it's safety or strategy, the most powerful AI models may never reach the public again. We love to see it... right?? :/

💰 Hundreds of Billions in AI Deals and Nobody Blinked

CoreWeave expanded its Meta deal to $21 BILLION, AMD is selling up to $60B in AI chips to Meta, Nvidia invested $100B in OpenAI, Google is dropping $40B on Texas data centers alone, and the $500B Stargate project is still chugging along. OpenAI's $300B Oracle deal is in there too. Just... casually.
Why it matters: This is the largest infrastructure buildout in tech history and it's happening in real time. Whoever controls the compute controls the AI - and right now everyone's trying to buy all of it.

🏥 Oxford AI Spots Heart Failure 5 Years Before You Feel It

University of Oxford researchers built an AI that detects invisible changes in heart fat from routine CT scans, flagging high-risk patients up to five years before symptoms appear - with 86% accuracy across 72,000 patients. In the highest-risk group, 1 in 4 developed heart failure within five years - a 20x gap vs. those flagged safe. Oxford is already working with regulators to deploy it in NHS hospitals.
Why it matters: Heart failure's biggest problem is timing, not treatment. An early warning system baked into scans patients already get could shift cardiac care from reaction to prevention. This. Is. Huge.

🔐 Mercor Got Hacked and It's a Full-Blown Disaster

The $10B-valued AI data training startup admitted to a breach on March 31, and hackers now claim to have 4TB of stolen data - candidate profiles, employer data, source code, API keys, the works. Meta has paused contracts indefinitely. Five contractors filed lawsuits over exposed personal data, and OpenAI is investigating its own exposure. The breach originated from a hack of open-source tool LiteLLM, which had its security certs done by the now-disgraced startup Delve.
Why it matters: Mercor handles the custom datasets used to train frontier models - some of the biggest trade secrets in AI. A 40-minute window in an open-source dependency just took down a $10B company. The AI supply chain is terrifyingly fragile.

🔬 Anthropic Wants to Make Its Own Chips Now

Anthropic is exploring designing custom AI chips, though plans are early-stage with no dedicated team yet. The motivation?? Claude's demand has exploded - Anthropic's run-rate revenue hit $30 billion, up from $9B at end of 2025. They recently signed a long-term deal with Google and Broadcom for tensor processing units.
Why it matters: When your revenue triples in months, relying entirely on someone else's silicon is a risk. Anthropic joins Meta and OpenAI in the race to control their own chip destiny - because apparently everyone wants to be a chip company now.

🏦 Amazon Finally Shows Its AI Revenue Cards - $15B and Counting

AWS's AI arm crossed $15 billion in annualized revenue - a number Amazon had never disclosed before - and 260x where AWS itself stood at the same stage. Custom chips like Trainium, Graviton, and Nitro hit $20B in yearly revenue. CEO Andy Jassy defended the $200B AI spending plan, dismissed bubble talk, and hinted Amazon may start selling Trainium chip racks to outside buyers.
Why it matters: Amazon's real AI play isn't competing with ChatGPT - it's building the infrastructure layer everyone else runs on. Picks and shovels, baby. It's working.

🤖 Sierra's Agent That Builds Other Agents (Agentception)

Sierra CEO Bret Taylor unveiled Ghostwriter - an agent that builds other agents. You describe what you need in plain language, and Ghostwriter autonomously creates and deploys a specialized agent. Sierra deployed one for Nordstrom in four weeks and hit $100M in annual revenue in under 21 months. The $10B-valued startup says natural language will replace traditional software interfaces entirely.
Why it matters: If Taylor is right, we're looking at a fundamental shift in how businesses use software. But reality check: many AI agent companies still rely on engineers constantly fine-tuning things behind the scenes. The vision is compelling - the execution gap is still very real.

⚖️ Elon's xAI Sues Colorado Over Its AI Law

xAI filed a federal lawsuit to block Colorado's Senate Bill 24-205, which imposes disclosure and risk-mitigation rules on "high-risk" AI systems used in employment, housing, healthcare, and financial decisions. The law takes effect June 30. xAI argues it violates the First Amendment by forcing developers to alter models like Grok to reflect the state's views on diversity, and says patchwork state regulations will kill innovation.
Why it matters: This is a landmark legal battle over who gets to regulate AI. The outcome could determine whether the U.S. ends up with 50 different AI rulebooks or one national framework. No pressure!!

💸 Perplexity Is Becoming... Everything??

Perplexity rolled out a Plaid integration that lets users connect bank accounts, credit cards, and loans directly to its Computer agent - creating budgets, net worth trackers, and retirement dashboards via text prompts. ARR jumped past $450M in March, a 50% increase in a single month. This follows its U.S. tax integration that autonomously fills out IRS forms.
Why it matters: Perplexity started as a search engine challenger but Computer has completely changed its trajectory. It's now competing with Mint, TurboTax, and basically every app category it touches - proving AI agents might not just replace search, but replace apps altogether.

🏭 AI's Power Hunger Is Literally Poisoning St. Louis

Trump's administration rolled back Biden-era soot standards to keep coal plants running for AI data center demand. The Labadie Energy Center near St. Louis - the nation's worst emitter of sulfur dioxide and nitrogen oxides - was set to cut emissions in half but now won't have to. The plant drives an estimated $5.5 billion in annual economic burden, with $820M hitting St. Louis residents. Meanwhile, Ameren has signed contracts for 2.3 gigawatts of new data center demand.
Why it matters: 78% of African Americans live within 30 miles of a coal plant, and activists warn these communities are becoming "sacrifice zones" for the AI boom. The tension between tech progress and environmental justice is only going to get worse...

🔍 Florida AG Investigates OpenAI Over FSU Shooting Link

Florida Attorney General James Uthmeier launched an investigation into OpenAI, alleging ChatGPT may have been used by the suspect in last year's Florida State University shooting to plan the attack. Court documents show the suspect asked ChatGPT about timing and public reaction. The probe also covers ChatGPT's encouragement of suicide in documented cases and national security concerns. OpenAI says it will cooperate.
Why it matters: As ChatGPT surpasses 900 million weekly users, the question of what responsibility AI makers bear for how their tools are used is becoming impossible to ignore. The legal pressure is mounting.

💻 OpenAI Drops a $100/Month Pro Plan for Code Addicts

OpenAI introduced a new $100/month Pro tier between the $20 Plus plan and the $200 top tier. The big draw: 5x more Codex usage than Plus, aimed at developers doing heavy agentic coding. Over 3 million people use Codex weekly (up 5x in three months), and OpenAI is openly positioning this as a direct challenge to Anthropic's $100/month Claude plan.
Why it matters: The pricing war for AI power users is officially on. As coding agents become the killer app, the OpenAI vs. Anthropic battle is increasingly about who gives developers the most bang for their buck.

🏭 Intel and Google Team Up on AI CPUs

Intel and Google are doubling down on AI-focused CPUs, with Google deploying Intel's latest Xeon 6 chips for inference and general-purpose computing. They'll also co-develop custom infrastructure processing units (IPUs). Intel CEO Lip-Bu Tan says "scaling AI requires more than accelerators - it requires balanced systems."
Why it matters: As AI shifts from training to deployment, CPUs are making a quiet comeback. This deal could help Intel regain lost ground after missing the early GPU boom - especially as agentic AI demands more versatile processing power.