🔥 Molotov cocktails & frontier models

🔥 Someone Firebombed Sam Altman's House

Daniel Moreno-Gama, 20, threw a Molotov cocktail at Sam Altman's home at 3:45 am, then drove to OpenAI's HQ threatening to burn it down. He'd published essays warning AI would end humanity and used PauseAI's Discord under the handle "Butlerian Jihadist." A second attack followed Sunday night - two suspects fired gunshots outside Altman's residence. Altman responded with a blog post calling AI anxiety "justified" and comparing the industry's power struggle to a "ring of power." PauseAI condemned the violence.
Why it matters: 4 in 5 Americans are now worried about AI, and anti-AI sentiment just went from online discourse to literal arson. Altman has become the face of every fear people have about this technology - and that target isn't shrinking.

🚨 Treasury Secretary & Fed Chair Summon Bank CEOs Over an AI Model

Scott Bessent (Treasury Secretary) and Jerome Powell (Fed Chair) called an urgent meeting at Treasury headquarters, pulling in CEOs from JPMorgan, Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley - all because Anthropic's new Mythos model is too good at finding security vulnerabilities in financial systems. It wasn't even trained for cybersecurity. Anthropic is now limiting access. U.K. regulators are also discussing the risks. The irony?? Anthropic is currently suing the Trump administration over the Pentagon labeling it a supply-chain risk.
Why it matters: When the Treasury Secretary and the Fed Chair personally call in bank CEOs over a chatbot, we have officially entered a new era. Mythos is proof that frontier AI models can be both a shield and a weapon for critical infrastructure - and nobody's quite sure which one it'll be.

👑 At HumanX, Everyone Was Talking About Claude - Not ChatGPT

At the HumanX AI conference in San Francisco, vendors and panelists kept naming Claude as the most popular chatbot. One vendor straight up said ChatGPT had "fallen off." Anthropic got shoutouts across panels while OpenAI's recent moves - ads in ChatGPT, Trump administration ties, a brutal New Yorker profile of Altman - have tanked the vibes. OpenAI fired back with a $100/month ChatGPT Pro tier with expanded Codex access, clearly aimed at Claude Code users. Board chairman Bret Taylor defended Altman, calling him "remarkable." Financially both companies are neck and neck - "the fastest-growing businesses in the history of tech."
Why it matters: OpenAI's perception problem is very real. Shedding Sora, adding ads, political entanglements... the brand is muddied. Claude's rise proves that product quality and focus can beat name recognition - even against a $122 billion juggernaut. The AI chatbot race is no longer a one-horse show.

🐴 Alibaba Was Secretly Behind "HappyHorse" - the Mystery AI Topping Video Rankings

Alibaba revealed it's the company behind "HappyHorse," a previously unknown video AI model that debuted at #1 on global rankings, knocking ByteDance's Seedance 2.0 to second place. Nobody knew who made it until after it had already won.
Why it matters: Alibaba just dropped a top-ranked video model out of nowhere under a codename. China's generative video race is moving FAST - and apparently the next breakthrough can just... appear on a leaderboard with a silly horse name.

🧬 AI Mined 400K Reddit Posts to Find Drug Side Effects That Clinical Trials Missed

Penn researchers fed over 400,000 Reddit posts about Ozempic and Mounjaro into GPT and Gemini using a technique called "computational social listening," mapping user reports to standardized medical terms across 5+ years of discussions. Nearly half the sample reported at least one side effect - flagging menstrual irregularities, chills, hot flashes, and fatigue that aren't on current drug labels and barely showed up in clinical trials. Researcher Lyle Ungar compared Reddit to a "neighborhood grapevine" of real-time patient notes.
Why it matters: FINALLYYYY an LLM use case that's genuinely useful. Reddit isn't a peer-reviewed journal, but when thousands of people flag the same symptoms, that signal is hard to ignore. AI just made it possible to listen to patients at scale - and pharma companies should be paying attention.

💰 TSMC Posts Its Fourth Straight Record Quarter

TSMC is expected to report a 50% surge in net profit to ~$17.1 billion for Q1 2026 - its ninth consecutive quarter of profit growth. Revenue already came in at 35% YoY growth, beating forecasts. Demand for their 3-nanometre chips and advanced packaging continues to outstrip production capacity. They're investing $165 billion in Arizona chip factories and upgrading Japan plans to include 3nm production. Market cap: roughly $1.6 trillion - nearly double Samsung's.
Why it matters: TSMC is the backbone of the entire AI hardware stack. As long as their order books overflow, it confirms that AI spending is not slowing down. Their earnings call Thursday will be the big signal for whether the AI capex boom has legs through 2026.

🇨🇳 China's Exports Lose Steam as Iran War Undercuts the AI Boom

China's export growth is forecast to slow sharply to 8.6% in March, down from 21.8% in Jan-Feb, as the Iran war and Strait of Hormuz closure spike energy and transport costs globally. Analyst forecasts are wildly split - Mizuho sees 24% growth, Citi projects just 3%. But South Korea's semiconductor exports to China still surged 151.4% on AI server demand, so the AI pipeline itself is still hot.
Why it matters: This is where geopolitics meets the AI supply chain head-on. The Iran conflict is stress-testing whether the global AI buildout can power through a major energy shock - and China's trade data this week will be one of the first concrete answers.

⛪ Anthropic Invited Christian Leaders to Discuss Claude's Soul

Anthropic invited Christian religious leaders to its San Francisco HQ for a summit on Claude's moral reasoning - covering grief responses, ethical frameworks, and whether AI could be considered a "child of God." The company, valued at $380 billion, is apparently taking the "alignment" part of AI alignment very literally.
Why it matters: Most AI companies consult ethicists and engineers. Anthropic is consulting theologians. Whether you find this inspiring or bizarre, no other major AI lab is publicly asking spiritual leaders to help shape their model's character. It's giving... existential crisis but make it corporate.

🏗️ Meta Poaching OpenAI's Stargate Executives

Meta is scooping up recently departed OpenAI Stargate executives - Peter Hoeschele, Shamez Hemani, and Anuj Saharan - to help build the new Meta Compute group and scale its AI infrastructure ambitions.
Why it matters: OpenAI's revolving door of departures continues to benefit its rivals. Zuckerberg is pulling senior infrastructure leaders from OpenAI's flagship data center project - the AI talent war shows no signs of cooling off.

🇨🇳 Chinese AI Startup StepFun Unwinds Offshore Structure for Hong Kong IPO

Shanghai-based StepFun, backed by state capital, Tencent, and Qiming Venture Partners, is dismantling its Cayman Islands corporate structure to comply with Beijing's crackdown on "red-chip" listings - clearing the path for a Hong Kong IPO at a $10 billion valuation. AI startup Moonshot (maker of the Kimi model) is also weighing the same move while raising $1 billion at $18 billion. Over 530 companies have filed to list in Hong Kong.
Why it matters: Beijing is tightening its grip on how Chinese AI companies structure themselves and raise money. The next wave of Chinese AI IPOs will have more government oversight and less founder flexibility. The rules of the game are changing.

🏢 OpenAI Opens a London Office, But Not the Data Center

OpenAI secured a permanent London office at Regent Quarter in King's Cross - capacity for 544 team members, opening in 2027. They currently have ~200 people in London. But here's the awkward part: just last week OpenAI paused its main UK data center project citing unfavorable regulations and high energy costs.
Why it matters: OpenAI is sending the UK a very mixed signal - "we'll hire your researchers but we're not building compute here." London gets the office but not the GPUs. The UK risks becoming a branch office rather than a power center.