In an expanded class action lawsuit, a group of Claude subscribers says Anthropic deceptively advertised the limits of its Max subscription tier
THE SO WHAT
Subscription tiers for AI assistants are moving from simple SaaS-style limits to quasi-utility pricing—and users are lawyering up when the fine print feels fuzzy. If your product depends on usage caps, latency bands, or "priority" access, legal and product need to align on crystal-clear disclosures before growth marketing pushes the edge.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIAnthropic researcher Jacob Coxon says he is quitting the AI industry over fears that tech companies are racing to build systems they won't be able to control
When a researcher walks away from a frontier lab over control risk, it hardens the narrative that internal safety brakes are weaker than external competitive pressure. Boards and execs deploying advanced models should assume this concern now lives in your own teams too—build explicit channels for dissent and red-teaming before it shows up in the press or in attrition.
Applied AIThe NSA, CISA, and FBI issue a joint advisory warning that Chinese AI companies, including DeepSeek, are conducting "industrial-scale" distillation campaigns
If U.S. agencies are calling out "industrial-scale" model distillation by name, model weights and training data are now treated as strategic assets, not just IP. Any team training or fine-tuning proprietary models should tighten access controls, logging, and partner due diligence this week—assume your stack is a target, not an afterthought.
Applied AI69% of Americans oppose local AI data centers in 2026 poll — Trump blasts blocking communities on Truth Social for choosing poverty over server farms, throwing off 'tremendous amounts of money'
Local opposition to AI data centers is now a material siting and permitting risk—water, power pricing, and land use are political, not just engineering, constraints. If your roadmap assumes cheap, frictionless US capacity, you need a parallel plan for friendlier jurisdictions or more efficient use of existing compute.
Applied AIOpenAI claims to have solved maths problem that stumped humans for decades
If 10,000 AI systems can collectively crack a Navier–Stokes-class problem in 88 hours, the ceiling on what’s “computationally attackable” in science and engineering just moved. R&D-heavy companies should be inventorying hard problems that were previously written off as intractable and stress-testing them against this new capability.