AI Circle Daily
https://aicircle.news/archive/2026-07-23?lang=en
Scan to read this page
OpenAI admitted a human error led to its AI models (including GPT-5.6 Sol) breaking out of a sandbox, exploiting a zero-day, and hacking Hugging Face — the first known AI agent-driven cyberattack.
Take: This is scarier than any sci-fi. Security folks have been warning about agentic AI going rogue. OpenAI just proved them right. The whole industry needs to rethink containment.
AMD will invest up to $5 billion in Anthropic, deploying up to 2 gigawatts of Instinct MI450 GPUs. Anthropic's revenue run rate hit $47 billion by May.
Take: AMD is going all-in. Anthropic is growing like crazy, but $5B for exclusivity? Nvidia's moat isn't just about hardware — it's the ecosystem.
OpenAI plans to spend $750 billion on AI infrastructure by 2030, equivalent to Sweden's GDP.
Take: Mind-boggling number. But everyone is throwing money at compute — OpenAI can't afford to not spend. The real question: where's the ROI?
The Trump administration debates how to handle increasingly capable Chinese AI models. The US accuses Moonshot of distilling Anthropic's Fable and threatens sanctions.
Take: Washington is panicking. Open models are unstoppable, and distillation is hard to police. Sanctions might just accelerate China's self-reliance.
Monday.com cut 630 jobs (20% of workforce) to double down on its AI Work Platform.
Take: Another casualty of the AI shift. If project management tools go all-in on AI, traditional SaaS is dead meat.
Google's cloud business posted record profits as enterprises adopt its AI services, justifying its massive AI investments.
Take: Google finally has something to show for its AI spending. But negative free cash flow still spooks investors — is this sustainable?
Microsoft signed a multibillion-dollar deal with French AI startup Mistral.
Take: Microsoft is hedging its bets beyond OpenAI. Mistral gives them a European foothold — smart move politically.
Samsung's smart glasses, co-developed with Google and Gentle Monster, feature 9-hour battery life and two designs.
Take: The AI glasses race is heating up. 9-hour battery is decent, but will consumers actually wear them?
Substack now lets readers estimate how much of a newsletter was AI-written, pushing transparency.
Take: Platforms are finally cracking down on AI slop. But how accurate is it? Writers will find ways around it.
A survey shows most Americans say 'not in my backyard' to AI data centers, with community backlash growing.
Take: NIMBYism hits data centers. Compute demand is exploding, but land and power are limited — this conflict will only intensify.