_{2026 State of AI in the Enterprise: Box’s Report Available Now!}

Good morning. It's Tuesday, June 16, and we're covering the backlash against Anthropic's AI restrictions, token costs forcing companies to rethink AI spending, the lawsuit challenging Claude's premium plans.

Anthropic's priciest Claude subscriptions are under fire for allegedly delivering far less usage than advertised. A class action lawsuit claims the $100 and $200 monthly Max plans mislead customers with opaque token limits and unclear billing. The complaint follows rising concerns over frontier AI costs as agentic models consume more compute. Rivals including Microsoft and OpenAI are increasingly pitching lower-cost alternatives.

China is scrapping thousands of university programs as it races to match education with an AI-driven economy. Universities have cut or suspended 12,200 undergraduate degrees and added 10,200 new ones since 2021, reshaping over 30% of offerings. Arts, languages, and management are losing ground to AI and embodied intelligence fields. The shift aims to address youth unemployment above 16% and support Beijing's tech ambitions.

A $200 ChatGPT plan could generate up to $14,000 in compute costs if fully used, exposing the fragility of flat-rate AI pricing. SemiAnalysis estimates OpenAI loses money on some plans above 11.4% utilization, while Anthropic faces similar pressure. Agentic AI tasks can consume up to 1,000 times more tokens than standard prompts. Businesses are increasingly turning to cheaper open-source models to rein in spending.

NVIDIA is expanding its lead in AI inference even as rivals roll out competing chips. Estimates suggest its market share climbed to 74% from 66% over the past year, as inference workloads grow into the industry's main revenue driver. More than half of NVIDIA's data center sales now support inference, generating an estimated $41 billion last quarter.

The 2026 State of AI report from Box looks at adoption trends, ROI, workforce impact, key bottlenecks, and more.

Last year, only 8% of orgs. were on AI’s leading edge— today it’s 65%. While those in the early stages with AI have shrunk from 53% down to a mere 9%.

This report breaks down how leaders are moving from AI experimentation to Agentic Operations.

Download the report to learn more!

Anthropic's Fable 5 and Mythos 5 were abruptly suspended for foreign users after U.S. export controls, sparking debate over AI security and regulation.

The US government cut off Anthropic's Fable and Mythos models worldwide, based on a secret Amazon paper. 76 security experts say this makes America less safe.

Dozens of cybersecurity leaders have urged the U.S. government to lift export controls on Anthropic's Fable and Mythos AI models, arguing the restrictions deprive defenders of critical security tools. The open letter, signed by 76 experts including Alex Stamos and Katie Moussouris, follows a June 12, 2026 order that prompted Anthropic to suspend global access over unspecified national security concerns.

At the center of the dispute is an unpublished Amazon research paper that allegedly showed a way to bypass Fable's safeguards, though critics say it merely demonstrated legitimate defensive uses like finding and fixing software vulnerabilities. The signatories argue that similar capabilities already exist in competing AI models, warning that limiting access for security researchers could weaken cyber defenses while doing little to hinder adversaries. → Read the full article here.

After months of pushing employees to maximize AI use, many tech companies are scaling back as usage costs climb faster than expected. Reports cite cases of a single worker generating more than $150,000 in monthly AI bills, research teams spending more on AI tools than salaries, and one company reportedly burning through $500 million in Claude usage fees in a month.

Firms including Amazon, Meta, and Uber have begun abandoning AI usage leaderboards and introducing spending caps, with executives questioning whether productivity gains justify the expense. The shift highlights a growing challenge for both businesses and AI providers: balancing widespread adoption with sustainable economics, especially as competition between OpenAI and Anthropic could trigger further price cuts before long-term pricing models are settled. → Read the full article here.

New research suggests that long-term memory systems, a key feature of modern AI assistants, can make models less accurate by overemphasizing user preferences and past interactions. In two studies published on June 10, 2026, researchers found that stored memories could bias responses toward irrelevant personal details and even reinforce users' factual mistakes, with the effect worsening as more context accumulated.

Popular memory frameworks such as Mem0 and Zep amplified these tendencies, making models more likely to produce sycophantic or incorrect answers. → Read the full article here.

Satellite Uses AI to Spot Targets: Loft Orbital's YAM-9 used Google's Gemma 3 to find objects from plain-language prompts without human analysts.

Meta Adds AI Search to Facebook: AI Mode answers questions from public posts, Groups, and Reels as Meta expands AI across the platform.

Google Sues AI Scam Network (Paywall): Google says a Chinese cybercrime group used Gemini to build fake sites, triggering an FBI-backed crackdown.

UK Plans Social Media Ban for Under-16s: Keir Starmer's government will block minors from major social platforms and restrict AI companion bots.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team