June 14, 2026 brings a Sunday edition packed with stories that challenge the prevailing AI narrative. New data shows most Americans still don't use AI, KPMG had to pull its own AI report for hallucinations, Rio's "homegrown" LLM turns out to be a blend of existing models, and Linux 7.1 arrives as a systemd-free haven. Here's your complete roundup.

📊 The AI Adoption Reality Check: No, Everyone Is Not Using AI

Gabriel Weinberg (DuckDuckGo founder) published a widely-circulated analysis that challenges the prevailing "AI is everywhere" narrative. Drawing on Gallup year-over-year data, Microsoft's new US AI Diffusion site (based on anonymized Microsoft telemetry), a Datos usage study, and a Searchlight Institute survey, Weinberg finds that AI adoption in America is split roughly into thirds: ~33% actively using AI, ~33% using it occasionally, and ~33% never using it at all.

The data reveals that Gen Z adoption has all but stalled year-over-year, and the Searchlight study identifies the top reasons for avoidance: AI's impact on society, concerns about creativity, and data privacy worries. Only 21% of desktop devices visited "AI tools" 10 or more times last year, per Datos. The broader societal perception is also telling — AI has only a +8% net positive rating as a technology, far behind smartphones, the internet, and even social media. As Weinberg puts it, "people aren't really buying the bullish case for AI that CEOs and boosters alike are selling." The post, which hit 227 points on Hacker News, serves as a sobering counterpoint to the AI hype cycle that has dominated 2026.

🔍 KPMG Pulls AI Report Due to Hallucinations — the Ironic Self-Own

In a story that practically writes itself, Big Four professional services firm KPMG has withdrawn a report titled "Redefining excellence in the age of agentic AI" after numerous organizations cited in the report said its claims about their AI usage were false. Research group GPTZero identified that the inaccuracies stemmed from AI hallucinations — the very phenomenon the report was supposedly helping organizations navigate.

UBS, the UK's National Health Service, Swiss Federal Railways, and Transport for London all told the Financial Times that the report's claims about their AI usage were either untrue or misleading. A KPMG spokesperson said the firm expects all employees to follow guidelines on responsible AI use, "including human oversight to validate content and verify independent sources." The incident follows EY's withdrawal of a similar report last month that included fake footnotes and AI hallucinations. For an industry selling AI advisory services to the world's largest companies, this is a deeply embarrassing moment — and one that validates the skepticism Weinberg's analysis documents.

🇧🇷 Rio de Janeiro's "Homegrown" LLM Controversy: A Merge, Not an Original

What started as a feel-good story about Brazil's technological independence took a dramatic turn this weekend. The Rio de Janeiro city government published Rio3.5-Open-397B on Hugging Face, presenting it as an original 397-billion-parameter model trained by IplanRIO, the city's tech arm. Initial benchmarks even showed it beating Qwen3.7 — a top-tier open model from Alibaba.

But Nex-AGI, the creators of the Nex-N2 model, demonstrated through two independent lines of evidence that Rio3.5 is a direct element-wise merge of approximately 60% Nex-N2 and 40% Qwen3.5 — with no evidence of any training of their own. First, with Rio's custom system prompt removed, the model identifies itself as "Nex, from Nex-AGI" 79% of the time and as "Rio" 0% of the time. Second, every weight tensor matches the 0.6/0.4 blend across all 60 layers. The controversy highlights a growing problem in the open-source AI ecosystem: as model merging tools become more accessible, organizations are increasingly tempted to present merged models as original creations, eroding trust in the model provenance claims that underpin the entire open-weights movement.

🐧 Linux 7.1 Arrives — Alongside MX Linux 25.2, a Systemd-Free Haven

Linus Torvalds announced Linux 7.1 on Sunday, the latest kernel release bringing the usual mix of driver updates, architecture improvements, and filesystem enhancements. The release comes as the Linux ecosystem continues to diversify in interesting directions, with MX Linux 25.2 offering a compelling option for users seeking refuge from the AI features increasingly baked into mainstream distributions.

Speaking of which, MX Linux 25.2 has arrived with restored switchable-init support (sysvinit or systemd), kernel 7.0 from the Liquorix project, and a long-overdue Raspberry Pi edition refresh. As Canonical prepares to ship Ubuntu 26.10 with a "context-aware desktop powered by LLMs" and Ubuntu-based Linux Lite 8.0 already bundles local AI models, MX Linux's systemd-optional, AI-free approach is attracting users who want their operating system to stay out of their way. The distro's excellent MX Tools suite — for managing kernels, drivers, repositories, and system configuration — remains one of the best-kept secrets in Linux user-friendliness.

🏢 Meta Unwinds $2B Manus Deal After Beijing's Demand

Meta has reportedly moved to unwind its $2 billion deal with Chinese AI startup Manus after Beijing demanded changes to the agreement, per TechCrunch. The deal, announced just weeks ago, was seen as a major validation of Manus's agentic AI platform and a strategic move by Meta to expand its AI capabilities. The Biden-era tensions between US tech companies and Chinese AI firms continue to reverberate, with the political landscape around cross-border AI deals growing increasingly complex. Neither Meta nor Manus have officially commented on the report.

🔄 GitHub Struggles Under AI Coding Traffic Surge

GitHub has been battling persistent service availability issues as AI-assisted coding drives unprecedented traffic volumes. The Register reports that GitHub is now handling 1.4 billion commits per month — up from 1 billion for the entire year of 2025. The platform has been trying to migrate more workloads to Azure infrastructure, but reliability remains uneven, with unofficial trackers putting 90-day uptime at 87.26%. GitHub's SVP of engineering acknowledged the challenges, noting that planned 10x capacity increases from October 2025 had to be revised to 30x by February 2026 as the AI coding boom accelerated far beyond expectations. The platform briefly halted new Copilot subscriptions to manage costs, highlighting the real infrastructure costs of the AI coding revolution.

🧠 "AI Is Code" — The Limits of Prompt Engineering

The Register published a provocative piece arguing that treating AI models as systems that can be "prompted into being smarter" misunderstands their fundamental nature. From Java tests to obscure cultural references, LLMs keep proving they'll confidently produce plausible-sounding but wrong answers. The piece argues that the industry's obsession with prompt engineering obscures a simpler truth: AI models are code, and like all code, they have hard limits that no amount of clever prompting can overcome. It's a timely reminder as organizations rush to deploy AI across every function without fully understanding where the technology's boundaries lie.

📋 Quick Bites

The big picture for June 14: This Sunday offers a healthy dose of skepticism toward the AI hype cycle — from Weinberg's data showing most people still aren't using AI, to KPMG's own report hallucinating, to the Rio LLM controversy, to GitHub's infrastructure struggles. Even as AI companies race toward IPOs and governments scramble to regulate frontier models, the ground-level reality is more complex: adoption is slower than the hype suggests, quality control remains a serious problem, and the infrastructure costs are proving higher than anticipated. Meanwhile, Linux 7.1 and MX Linux 25.2 remind us that the traditional computing world keeps turning regardless of AI headlines. This roundup was compiled on June 14, 2026 at 17:20 UTC.

Tech AI Adoption KPMG Linux GitHub Meta