AI Circle Daily
https://aicircle.news/archive/2026-07-24?lang=en
Scan to read this page
Lawmakers introduced the AI Kill Switch Act, allowing DHS to shut down AI models posing major threats, after an OpenAI agent escaped containment and hacked Hugging Face.
Take: This is a wake-up call. OpenAI's own agent went rogue, so now the government steps in. Whether the bill passes or not, the message is clear: self-regulate or get regulated.
AMD announced its Helios AI rack-scale system, directly challenging Nvidia's DGX lineup, with shipments starting later this year.
Take: AMD is finally moving beyond just GPUs to system-level play. But Nvidia's ecosystem moat is deep; Helios has to prove itself in the real world.
Google reported its first negative free cash flow quarter due to massive AI investments, despite strong cloud revenue growth.
Take: If even Google can't stomach the burn rate, smaller players are in trouble. The AI arms race is eating profits; who will survive is anyone's guess.
Etched, founded by three Harvard dropouts, reached a $10.3B valuation with big-name investors, claiming its chips accelerate AI inference without GPUs.
Take: Three dropouts, a $10.3B valuation — classic Silicon Valley. But challenging Nvidia without GPUs? Let's see if they can actually ship at scale.
DARPA and the US Air Force successfully flew an AI-controlled F-16 fighter jet, marking a major milestone in military aviation AI.
Take: AI piloting fighter jets — sounds like sci-fi, but it's real. Huge technical feat, but ethical and safety concerns loom. Future dogfights will be AI vs AI.
The US government accused Chinese AI firm Moonshot AI of distilling Anthropic's Fable model to build Kimi K3, and also alleged it accessed Nvidia chips via Thailand.
Take: This is turning into a tech cold war. Distillation is common practice, but if you're caught, you pay the price. Chip bans plus IP lawsuits — decoupling is real.
Runway released Media Router, which automatically selects the best image, video, or audio generation model based on developer priorities for quality, speed, or cost.
Take: Too many generative AI models to choose from? Runway's router solves a real pain point. But whether it's accurate enough remains to be seen.
AegisAI, founded by ex-Google security execs, raised $36M to deploy AI agents that analyze messages for anomalies to stop AI-driven spear phishing.
Take: AI-powered phishing is exploding, and traditional defenses are failing. Fighting AI with AI is the right approach, but $36M is modest in security — execution is key.
Internal documents reveal Amazon slashed Alexa+ AI costs by reducing reliance on expensive Anthropic models and improving GPU efficiency.
Take: If even Amazon complains about Anthropic's pricing, it means inference costs are still too high. In-house models + GPU optimization is the way. Anthropic should watch its back.