Claude is down as Anthropic investigates errors across its services
THE SO WHAT
High error rates across Claude’s web, code, cowork, and API products show how vertically integrated AI suites share a single point of failure. If you’ve standardized on one vendor’s stack, document which workflows hard‑fail during outages and where a lighter backup model would be "good enough."
READ THE SOURCE
MORE FROM THE WIRE
Applied AIFrom event posters to cafe menus, AI slop is everywhere in the real world
When low-stakes surfaces like posters and menus are saturated with low-quality AI, it trains customers to tune out entire visual channels. Brand and product teams should decide explicitly where they tolerate AI generics — and where they still pay for human craft because attention is scarce there.
Applied AIAI needs $6tn a year by 2031 to pay for its data centres, Bain says
If Bain is right that current data center build-out implies ~$6T in annual AI revenue by 2031 versus $1.2T–$1.8T from existing products, then a lot of infra is being built on unproven demand. Operators should pressure-test their own AI revenue assumptions against this gap — and avoid locking into fixed-cost footprints that only pencil out under best-case adoption curves.
Applied AIOpenAI pulls new AI model release following widespread security worries
Pausing GPT-6.1 Astra weeks before launch over safety concerns shows that frontier model release is now gated as much by security posture as by capability. Enterprises building on vendor roadmaps should assume slippage and build abstraction layers — don’t tie critical launches to specific model versions or dates.
Applied AIBill Gates tells Ezra Klein AI crossed bio and cyber thresholds this year
If AI has crossed Gates’ “danger thresholds” for enabling bioterror and crippling cyberattacks, then advanced models are now squarely in the national security and critical infrastructure domain. Boards should treat access to high-end models, code, and data the way they’d treat dual-use tech — with export, insider risk, and misuse scenarios formally on the risk register.