AI news · Friday, August 28, 2026
Anthropic legal win highlights growing friction over AI national security
Anthropic just secured a significant court victory against the Pentagon after a federal judge ruled that designating the company a 'supply-chain risk' was illegal and amounted to political retaliation. The conflict stemmed from the company’s refusal to build safety overrides specifically for autonomous weapons and surveillance systems. While Anthropic is busy winning in court, it’s also making waves in R&D: researchers just published a paper on 'Automated Alignment Researchers' that can improve model safety benchmarks better than humans, and at a fraction of the cost—$4 an hour versus $150.
Meanwhile, the industry’s hunger for 'open-weight' models—software you can download and run yourself—is triggering a shopping spree. Nvidia is reportedly deep in talks to buy Hugging Face for $13 billion to secure its position as the de facto home for open-source AI development. This follows a trend of tech giants snapping up model builders, like Stripe’s recent $7 billion grab of OpenRouter.
Companies are turning to these open models because they’re cheaper for repetitive tasks like customer service, which is a major draw as the AI 'Great Flattening' leaves managers feeling the burn of high inference costs and larger, under-managed teams. Speaking of money, Meta is attempting to close the book on teen safety litigation by settling for $18 billion over the next decade. The kicker: the settlement only stays at that level if competitors like TikTok and YouTube also agree to similar restrictions.
It’s a clever bit of leverage that costs Meta less than 2% of last year’s profit. While big companies play legal chess, Google is making its search results even more annoying by forcing its AI summaries to auto-expand for certain queries, pushing traditional links further down the page. Hardware enthusiasts are having a weird week too.
Modders have leaked Nvidia’s unreleased DLSS 5 'neural rendering' code, effectively letting them force AI-enhanced textures into games like GTA V, though it's causing VRAM spikes and mixed reactions on character visuals. Even consumer gadgets are getting the AI treatment, though with mixed results: a $300 set of smart glasses can technically play movies, but mostly just acts as a warm, redundant screen for the laptop already sitting in front of you. Perhaps most tellingly, even the early AI skeptics are breaking: Minecraft creator Notch is now 'vibe coding' with AI after admitting it was time to 'eat some crow' because he couldn't find human programmers fast enough.
The quick hits
- A federal judge ruled the Pentagon’s blacklisting of Anthropic as a security risk was illegal — a major win for the startup in its battle over AI safety protocols.
- Nvidia is reportedly in talks to buy Hugging Face for $13 billion — a move to control the ecosystem where most open-source AI is built and shared.
- Meta settled a teen safety lawsuit for $18 billion — a massive sum that is actually a bargain when spread over a decade of payments.
- Google is auto-expanding its AI search summaries — making users scroll past even more text to reach the actual websites they searched for.