Editorial note

Today’s thread ties two threads: invisible surveillance riding ordinary tech, and the growing pains of AI moving from hype to operational reality. Small changes — a firmware tweak, a public audit, a removed Excel function — matter because they change what’s visible, what’s trusted, and what you need to do.

In Brief

Microsoft Is retiring Excel’s COPILOT() function

Why this matters now: Microsoft’s removal of the =COPILOT() worksheet function means Excel workbooks that use plain‑English AI prompts in cells will stop working on September 14, 2026 unless users migrate them to the Copilot side pane.

Microsoft says the inline formula was a preview feature and that the same AI features remain in the Copilot side pane, but warns the cell function was never intended for precise or compliance‑sensitive calculations. If your team uses Copilot functions in shared workbooks, now is the time to audit and replace those calls or move outputs to supported interfaces. As one common reaction put it: “audit those sheets now” — this is a short, urgent ops task for finance teams and spreadsheet power users. See the company notice in the original post.

“Those capabilities remain available through the Copilot side pane.”

Anthropic CEO: curing cancer is how AI wins public trust

Why this matters now: Anthropic CEO Dario Amodei argued that tangible health breakthroughs — for example a real cancer treatment enabled by AI — would do far more to change public opinion than marketing or slogans.

Amodei framed the conversation as a reality check: AI companies have “overpromised and undersold,” and regulators or watchdogs are reasonable to demand demonstrable benefits. That shifts the public debate from abstract fear of models to concrete evaluation of outcomes in healthcare and other high‑stakes fields. The full comments are summarized in the thread.

Deep Dive

Ordinary Wi‑Fi can now identify you with near‑perfect accuracy

Why this matters now: Karlsruhe Institute of Technology researchers demonstrated a method that uses routine Wi‑Fi beamforming data to create radio‑based “images” of a room and identify people with up to ~99.5% accuracy — and the attack works with off‑the‑shelf hardware.

The research — presented as a demo dubbed the BFId attack — leverages unencrypted protocol signals that routers and devices already exchange while aligning antennas and steering beams. Those beamforming frames can be passively recorded and processed to infer movement patterns, presence, and identity without the target carrying a phone or wearing a tag. The core technical point is simple: modest‑frequency radio reflections encode rich spatial information, and modern machine learning can map those signatures back to individuals.

This is not a sci‑fi, one‑in‑a‑decade exploit. The vulnerability stems from routinely broadcast, unencrypted control data in Wi‑Fi standards. That matters because fixes aren’t just a software patch on one brand’s router: they likely require firmware updates across vendors or adjustments to protocol standards to stop leaking identifying beamforming metadata. Practically, that means homeowners and IT teams face a delayed window of exposure where equipment on shelves today can be exploited tomorrow.

There are caveats. The team’s demo was done in controlled settings; real‑world performance depends on layout, concurrent Wi‑Fi traffic, and noise. Legal hurdles and operational complexity also limit rapid abuse compared with, say, a smartphone camera. Still, the attack cuts two ways: it shows a new surveillance vector for malicious actors and a potential sensor for useful, privacy‑preserving applications (smart homes that detect falls without cameras). The public reaction in the original thread split between alarm — “turn off your Wi‑Fi” — and calls for firmware fixes and standards work.

“By observing the propagation of radio waves, we can create an image of the surroundings and of persons who are present.” — Prof. Thorsten Strufe

What you can do now:

  • Update router firmware and network gear from vendors that commit to security patches.
  • Disable guest networks and unnecessary SSIDs where possible.
  • For sensitive environments, consider wired networks or faraday‑style attenuation for critical rooms until vendors address beamforming metadata.

HaveIBeenFlocked: a searchable ledger of license‑plate searches

Why this matters now: HaveIBeenFlocked.com assembled FOIA‑obtained audit logs from roughly 6,586 U.S. agencies to let drivers see whether their license plate was searched in Flock Safety’s camera network — exposing how often and why plates are being queried.

Flock Safety runs over 120,000 AI‑powered license‑plate readers and shares searchable lists with thousands of law enforcement customers. The website aggregates redacted audit logs that governments released and shows the stated reason for each search — common reasons include theft and drug investigations — while warning that records are incomplete and often redacted. For people who care about location privacy, this is a rare window into how private corporate camera networks intersect with public policing.

The project’s value is twofold. First, it creates individual agency: people can check whether city or county systems captured them, and what rationale was recorded. Second, it provides empirical material for policy and litigation — watchdogs can now quantify search frequency, identify patterns of misuse, and press for retention limits, stricter access controls, or local bans. The creators told local media bluntly: “If your city is using Flock, you can search ‘Have I Been Flocked,’ … and see what searches are being done.” The thread captured both praise for the transparency tool and concerns about patchy records.

Important limitations: the dataset is not comprehensive. Some agencies refuse or heavily redact logs, and appearing in a log doesn’t necessarily mean you were the subject of an investigation. But as a transparency tool it lowers the cost of oversight and gives journalists and civil‑liberties groups material to work with.

Practical implications:

  • Municipalities using private camera vendors should expect public scrutiny and FOIA pressure.
  • Individuals in affected areas can check logs and raise questions with local law enforcement if they find unexpected searches.
  • Policy debates about surveillance will now have richer empirical evidence, tilting arguments from abstract harms to concrete datasets.

Closing Thought

Routine tech is increasingly doubling as a sensor and an accountability record. Ordinary Wi‑Fi and license‑plate feeds both show that when a protocol or vendor sits between people and public systems, privacy and oversight questions move from theory to immediate practice. Short‑term fixes matter — audits, firmware updates, and public records — but the bigger answer will be policy and standards that align vendor incentives with civil‑liberties expectations.

Sources