Border Patrol's predictive policing units are mining financial data on Americans and directing local cops to pull them over — with no specific crime suspected. 404 Media is naming the unit publicly for the first time. This is surveillance infrastructure being normalized in plain sight.
OpenAI used an unreleased model to resolve one of mathematics' seven Millennium Prize Problems — Navier–Stokes existence and smoothness. These problems have stood unsolved for decades. AI is now closing them faster than the math community can track.
Another safety-focused departure from a frontier lab, this time with explicit apocalyptic framing. This follows Anthropic's head of safeguards exit in February and OpenAI dissolving its own mission guardrails. The pattern of insiders walking out and warning loudly is accelerating.
Kimi's surprise capability jump that spooked markets wasn't purely homegrown — U.S. agencies say it was trained on American foundation models. The AI export control debate just got a lot more complicated.
License plate surveillance company Flock ran a webinar showing police how to monitor protests, parades, and public gatherings. The company's tools are already embedded in hundreds of local departments. The infrastructure for mass protest surveillance is already built and being actively taught.
Fields Medal winner Terence Tao says the pool of "good, fruitful open problems" is being mined non-renewably — even the rumor of someone working on a problem now triggers floods of AI-powered attempts. Math's frontier is being consumed faster than it can regenerate.
Mollick flagged that discourse from people inside the labs is "super weird, even for AI discourse" and that everyone appears to be in reaction mode to fast-moving developments. Coming alongside a Millennium Prize solution, a high-profile researcher resignation, and Kimi's provenance controversy, this reads less like routine anxiety and more like something significant is happening behind closed doors that hasn't fully surfaced yet.
Deutscher catalogued 2026's AI safety departures in a single post: Anthropic's head of safeguards quit in February warning "the world is in peril," OpenAI dissolved its mission guardrails, and now another researcher exits with Terminator-level warnings. Strung together like this, the list is striking. These aren't fringe voices — they're the people who were paid to know where the risks are.
The Kobeissi Letter noted that the AI Revolution is fueling a record corporate bond financing boom — U.S. high-grade issuance expected to hit ~$215 billion in September alone. Capital markets are pricing in an AI-driven growth supercycle even as safety insiders are quietly walking out. The gap between Wall Street's optimism and lab insiders' alarm has rarely been wider.
Two stories today that don't look related are actually the same story. OpenAI solves a Millennium Prize Problem using an unreleased model. Terence Tao warns that AI is strip-mining mathematics' best open problems faster than the field can replenish them. The capability and the consequence are arriving simultaneously — we're getting the answers before we've figured out what it means to let machines own the discovery.
And underneath all of it, the surveillance layer keeps quietly expanding. DHS is running financial-data-driven predictive policing. Flock is running webinars teaching cops to track protesters by plate. These aren't hypothetical futures — they're operational systems being refined in real time. The same week AI cracks century-old math problems, it's also being used to pull over people who've committed no crime. Both things are true. Both deserve the same front page.