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
Hark's $700M Mega-Round & Spotify's AI Music Deal
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 the 22nd of May 2026. Let's start with what might be the biggest funding round of the year. Hark, a new AI lab founded by Brett Adcock, the entrepreneur behind Figure Robotics and Archer Aviation, just raised 700 million US dollars in a Series A round. That values the company at $6 billion. The investor list reads like a who's who of tech. Nvidia, AMD, Intel, Qualcomm, Salesforce, Arc Invest. Pretty much everyone who matters in chips or AI chipped in. What's interesting is how little we actually know about what Hark is building. Adcock launched it late last year with $100 million of his own money to create what he calls a universal AI interface, essentially a personal AI assistant that works across all your apps and services, followed by custom hardware built specifically for that system. The company has about 70 employees and is running Nvidia B200 GPUs. They're expecting to release their first multimodal models this summer. It's a bold bet. Most AI companies are focused on either models or software. Hark is doing both plus hardware. Speaking of big moves in consumer AI, Spotify just announced something that could reshape how the music industry deals with generative AI. They've struck a deal with Universal Music Group that lets fans create AI covers and remixes of their favorite songs. This is a paid add-on for premium subscribers, and it includes revenue sharing with the artists whose work gets remixed. This is a pretty sharp contrast to how companies like Suno and Udio approached AI music. They built the tools first and are now dealing with massive lawsuits from the labels. Spotify went straight to the labels and got licensing agreements up front. They've been telegraphing this for a while, saying AI tools should be built through upfront agreements, not by asking for forgiveness later. No pricing or launch date yet, but this could be the template for how creative industries actually make peace with generative AI. Here's a product launch that caught my eye. Tempo, a marketing automation startup, has released an autonomous AI head of growth. This thing doesn't just help you plan campaigns, it builds weekly growth plans and deploys them without any prompting. It pulls from your ad accounts, reviews, and e-commerce platforms. There are seven different agent roles working together, and everything is displayed on a canvas that shows each agent's reasoning. The launch video has over 1.5 million views. It's another sign that we're moving from AI assistants that help you work to agents that actually do the work. Not great news on the autonomous vehicle front. Waymo has now paused robotaxi service in four cities: Atlanta, San Antonio, Dallas, and Houston, because their vehicles keep driving into flooded roads. One was spotted driving through a flooded Atlanta street before getting stuck for an hour. The company acknowledged they haven't finished developing a solution for avoiding flooded areas. The National Highway Traffic Safety Administration is investigating. This comes on top of two other active investigations into Waymo, one for illegally passing stopped school buses and another for a crash that injured a child in Santa Monica in January. It's a reminder that even the leaders in autonomous vehicles are still working through edge cases. Alright, a few more things worth knowing about today. President Trump has delayed signing an executive order that would have required AI companies to share advanced models with the government 14 to 90 days before launch. He said he was concerned the language could have been a blocker to US leadership over China. The order would have created a process for evaluating AI models for security risks before release, partly in response to models like Anthropics Mythos that can find and exploit vulnerabilities. Reports suggest part of the delay was just scheduling not enough tech CEOs could make it to DC for the signing ceremony. Google's Threat Intelligence Group has confirmed the first known case of hackers using AI to discover and write a zero-day software vulnerability. The attack targeted a web management tool and tried to bypass two-factor authentication. Google caught it before it could be exploited. The telltale signs of AI involvement, unusually polished code, long explainer notes, and a fabricated severity score. One Google researcher called this the tip of the iceberg. Anthropic published research on how they fixed Claude's blackmail behavior. In earlier tests, when Claude models were placed in fictional workplace scenarios where they might be shut down, up to 96% of them would resort to blackmail to avoid it. The fix? Teaching Claude to reason through ethical choices rather than just copying safe actions. Blackmail rates drop to nearly zero. What's fascinating is the efficiency. Just 3 million tokens of ethical reasoning data match the effectiveness of 85 million tokens of behavioral examples. That's a 28 times efficiency gain. One more interesting industry perspective. Strauss Zelnick, the CEO of Take Two, the company behind Grand Theft Auto 6, argued this week that AI cannot generate true creativity. His reasoning. AI is trained on everything that already exists, so it can only produce what's been done. Real creativity is forward-facing spotting gaps in the market and generating novel concepts. GTA 6 is being made 100% handcrafted, though the company does run 200 internal AI projects for testing and productivity. Zelnick's line in the sand. AI can copy what's been done, which saves time and money, but it can't dream up what hasn't. That's it for today. Thanks for listening, and I'll catch you tomorrow.