AI Mornings with Andreas Vig
Your daily AI news briefing in under 10 minutes. New models, product launches, research breakthroughs, and industry shifts, explained clearly, no hype.
AI Mornings with Andreas Vig
Altman's Home Attacked & US Government Endorses Anthropic's Mythos
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Hey, welcome to AI Mornings with Andreas Vig. It's April 13, 2026. Sam Altman had a rough Friday night. Someone threw a Molotov cocktail at his San Francisco home in the early morning hours. No one was hurt, and police arrested a suspect at OpenAI headquarters shortly after, where he was threatening to burn down the building. Altman connected the incident to a New Yorker investigative piece that had just been published. The article, written by Pulitzer winner Ronan Farrow and Andrew Marents, painted Altman as having, quote, a relentless will to power, and quoted sources questioning his trustworthiness. Altman responded with a blog post acknowledging he's made mistakes, including being too conflict averse, but defended his commitment to OpenAI's mission. He also warned about escalating rhetoric around AI development, saying we should have fewer explosions and fewer homes, figuratively and literally. It's a stark reminder of how heated the AI debate has become. Meanwhile, Anthropic's Mythos model is getting an unexpected endorsement from the US government. Treasury Secretary Scott Bissent and Federal Reserve Chair Jerome Powell reportedly summoned major bank executives this week to encourage them to use Mythos for detecting security vulnerabilities. That's notable for a few reasons. First, Anthropic announced Mythos this week and said they wouldn't release it publicly because it's too good at finding vulnerabilities. Second, Anthropic is currently suing the Defense Department over a supply chain risk designation. And third, this shows how quickly the cybersecurity capabilities of frontier models are being taken seriously at the highest levels. Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley are now testing the model alongside original partner JP Morgan. UK regulators are also discussing Methos risks. Perplexity is making a big pivot. The company built its name trying to compete with Google Search, but their computer agent is now positioning them against an entirely different set of competitors: Mint, TurboTax, and personal finance apps. Perplexity just launched a plaid integration that lets users connect bank accounts, credit cards, loans, and brokerage accounts. The agent can then build budgets, debt payoff plans, and retirement dashboards from simple text prompts. The move seems to be working. Perplexity's annual revenue run rate jumped 50% in March to hit 450 million US dollars. That's the power of agents that actually do things rather than just answer questions. In his annual shareholder letter, Amazon's CEO revealed that the company's AWS AI arm has crossed 15 billion US dollars in annualized revenue. For context, that's 260 times where AWS itself stood at the same stage of its development. Amazon's custom chips, including Trainium, Graviton, and Nitro, are now generating 20 billion US dollars yearly. Jassy hinted that Amazon might start selling trainium racks to outside customers, which would be a direct challenge to Nvidia's dominance. The letter came after investors were rattled by Amazon's planned 200 billion US dollar capital expenditure, but Jassy argued the spending is backed by locked-in customer demand. On the hardware front, Chinese robotics company Unitree is bringing a surprisingly affordable humanoid to Western markets. The R1 humanoid will be available in North America, Europe, Japan, and Singapore starting next week for 4,370 US dollars. That's a fraction of what other humanoids cost. It's designed for dynamic movements like cartwheels and downhill running. Unitree aims to ship between 10,000 and 20,000 units this year, which would be a serious test of whether affordable humanoids can break into mainstream Western markets. Apple is quietly working on smart glasses. According to Bloomberg's Mark German, the company is testing four different frame designs ahead of a planned 2027 launch. The designs range from large, rectangular frames to slimmer ones similar to what Tim Cook wears, plus oval and circular options. Unlike the Vision Pro, these won't have displays. They'll be more like Meta's Ray Bans handling photos, videos, calls, and music, plus that long-promised Siri upgrade. It's a pragmatic step back from Apple's ambitious mixed reality plans after Vision Pro's struggle to gain traction. In Health AI, researchers at Oxford have built something impressive. Their system reads CT scans and detects changes in heart fat texture that are invisible to human doctors, flagging patients at high risk of heart failure up to five years out with 86% accuracy across 72,000 patients. In the highest risk group the AI identified, one in four patients developed heart failure within five years, versus a tiny fraction for those it flagged as safe. Oxford is working with regulators to bring this to NHS hospitals and extend it to all chest CT scans. A couple of other things worth noting. ScienceR opened pre-orders for Loom, a floor lamp that transforms into a robot that folds laundry and makes beds, shipping this summer. And Clone Robotics laid out plans for synthetic humans by 2028, claiming they can manufacture full musculoskeletal androids for under 20,000 US dollars at scale, though that's still unproven at production scale. That's all for today. Catch you tomorrow.