Proton, the privacy-focused email company, is forcing employees through AI competency interviews where the outcome determines job security. A leaked internal email confirms these are "consequential" reviews — not optional training. The quiet part is now loud: AI adoption isn't a perk, it's a performance metric.
OpenAI released GPT-6.1 Terra, a fast and efficient model emphasizing visual reasoning and computer use. Willison notes the framing is telling — OpenAI is doubling down on agents that can see and act, not just chat. This is the architecture of a model built to use software, not just talk about it.
Reducto's CEO argues that the real bottleneck for AI agents isn't reasoning — it's document ingestion. PDFs, scans, and spreadsheets remain largely illegible to frontier models at production quality. Healthcare, legal, and finance can't go agentic until this layer is solved.
In leaked audio from a private police conference, Flock Safety CEO Garrett Langley dismissed community criticism of his license plate reader network as manufactured. He also shared that his wife told him never to apologize. Surveillance tech executives now appear to be coaching law enforcement on how to handle PR, not just crime.
Datavault AI, a company whose entire strategy rested on holding Bitcoin, halved in value after admitting it cannot access its own treasury. With Bitcoin near $80K and sentiment cautious ahead of the Fed decision, this is a spectacular stress test of the "Bitcoin treasury company" playbook going wrong.
Early adopters of the iPhone Duo are already reporting a visible crease down the center of the display. Dubbed "Crackgate," the issue arrives almost immediately after launch for a device that costs as much as a used car. Apple's foldable moment is off to a rough start.
Mollick flagged a significant research finding: experts using AI agents are being nudged toward the AI's positions even when they believe they're exercising independent judgment. This isn't a fringe user problem — it's happening to trained professionals. The autonomy gap between "I'm using AI as a tool" and "the AI is shaping my conclusions" is narrowing fast, and most people don't notice when they cross it.
Databricks CEO Ali Ghodsi made a striking claim: even if frontier model progress stopped today, most enterprises would still be bottlenecked — not on model capability, but on context. The implication is that the competitive moat is shifting from who has the best model to who has the richest proprietary data layer feeding their agents. Commoditized intelligence, differentiated memory.
Artificial Analysis released Intelligence Index v5, moving to closed benchmarks with private datasets. The finding: open-weight models are falling further behind frontier models than previously measured on saturated public evals. The gap between what you can run locally and what you can access via API appears to be widening again, just as the narrative of open-source parity was gaining momentum.
Two stories from today share a structural DNA even though they look nothing alike. Proton re-interviewing staff on AI competency and Flock's CEO privately dismissing public backlash are both, at root, about institutions deciding that inconvenient feedback — from employees worried about their jobs, from communities worried about surveillance — doesn't deserve a real response. The answer in both cases is internal messaging, not course correction.
Ethan Mollick's finding that even experts are being unconsciously steered by AI agents should be read alongside the Flock story and the Proton story. We are building systems optimized to act, and we are simultaneously discovering that the humans supervising those systems are more suggestible than assumed. That's not a coincidence — it's the operating condition of the next decade.