Today at TechScan: Retro‑hacks, Agent Wars, and the Cost of Tooling

The most consequential tech story today isn’t a shiny model launch or a new gadget teardown—it’s a reminder that some of the internet’s most relied-on infrastructure still runs on paperwork, platform whim, and a single “verified” account. Microsoft abruptly terminated the account used by VeraCrypt developer Mounir Idrassi (and his organization, IDRIX) to sign Windows drivers and the bootloader, effectively halting Windows updates for one of the best-known open-source disk encryption tools. Idrassi says there was no warning and only a terse message that verification failed, with no appeal and no public explanation from Microsoft. If you’ve ever wondered how “secure by design” collides with “support queue by automation,” this is the collision in slow motion—and it lands on end users who did nothing wrong except depend on a widely used project that must pass through a single corporate gate to ship trusted binaries on Windows.

What makes this episode feel bigger than a single project is the shape of the dependency. Code signing isn’t a vanity badge; for drivers and bootloaders it’s table stakes for distribution and functionality. VeraCrypt can still exist as code, but its ability to deliver safe, signed updates to the platform where “most VeraCrypt users” live is suddenly constrained by an opaque verification process. The 404 Media reporting notes WireGuard has described similar sudden suspensions, which hints at a pattern: even well-regarded security software can be throttled by account-level enforcement that may be automated, under-explained, and hard to contest. We spend a lot of time talking about supply-chain security as if it’s purely a cryptographic or procedural problem; today’s mess is the administrative layer failing like a single brittle link.

That fragility also rhymes with another trend across today’s briefing: the growing gap between what’s technically possible and what’s operationally survivable. Whether you’re building an encryption tool that needs a signing account, or an “agent platform” that needs predictable billing and governance, the limiting factor increasingly isn’t the cleverness of the technology—it’s the reliability of the systems around it: verification workflows, distribution channels, and institutional trust.

The surveillance business is wrestling with a different version of the same problem: who controls the chokepoints, who gets access, and what happens when governance is an afterthought. CNET reports that dozens of U.S. cities and jurisdictions have suspended or ended contracts with Flock Safety in 2026 amid public backlash over data privacy and the company’s expanding surveillance offerings. Flock is known for AI-powered automated license plate readers, and it’s not staying in one lane: the company also sells advanced street cameras and even a “Drone as First Responder” system that can launch drones to track vehicles or people. The technical pitch is public safety; the political reality is that the data has a habit of traveling farther than residents expect.

A key accelerant in the cancellations is reporting that law enforcement agencies sometimes share Flock data with federal agencies like ICE, triggering backlash and terminations. CNET also notes Ring/Amazon cut ties after planned integrations drew criticism, a telling detail: when surveillance products start behaving like platforms—integrating, federating, and cross-pollinating datasets—local procurement decisions stop feeling local. The objections become less about any single camera and more about the emerging mesh: data sharing practices, unclear limits, and oversight that struggles to keep up with “just one more integration.”

At the federal level, NPR adds another hard-edged data point to the surveillance debate: ICE has acknowledged it is using spyware called Graphite, including zero-click capabilities that can access encrypted messages, according to a letter from Acting Director Todd Lyons. The agency frames the use as part of Homeland Security Investigations’ efforts to disrupt fentanyl trafficking, and says it will comply with constitutional requirements and involve legal advisors. But the disclosure lands in a landscape already inflamed by reports from WhatsApp and Citizen Lab that Graphite targeted journalists and civil-society figures. Lawmakers pressing ICE want details on who can be targeted and the legal basis for domestic use—exactly the kind of questions that tend to arrive after the tooling is already in motion.

If that sounds like a lot of institutional power concentrated in a few tools and contracts, today’s maker and retrocomputing stories offer a counterpoint: individuals rediscovering power by working beneath the glossy abstractions. One hobbyist has successfully ported Mac OS X 10.0 “Cheetah” to the Nintendo Wii, and the reasons it worked are almost a history lesson masquerading as a build log. The project adapts the open-source Darwin/XNU core to the Wii’s PowerPC 750CL CPU—hardware that sits close enough to G3-era Macs to be plausible, but far enough away in boot and I/O expectations to demand real engineering. The author validated RAM sufficiency with QEMU, then did the work vendors usually hide behind firmware: building a custom bootloader, patching the kernel, and replacing the behaviors normally handled by Mac Open Firmware and BootX.

The nitty-gritty here is the point, not the novelty. The port required serial debug support, SD-based booting, interrupt and framebuffer handling, and USB input support—an entire boot-and-driver stack assembled to coax a closed-source userland into breathing on alien hardware, made feasible because Darwin is open enough that, once the kernel and I/O are handled, the rest of macOS components can run. The Hacker News discussion highlights how the write-up doubles as a practical tour of reverse-engineering and systems tooling—people talking about Hopper, Ghidra, and IDA—because “running macOS on a Wii” is really shorthand for “remember how computers actually start.”

That same theme—real engineering hiding behind product decisions—shows up in a teardown of LG’s unreleased Rollable phone prototype, covered by Ars Technica. JerryRigEverything’s disassembly reveals the kind of intricate machinery that makes rollables a demo-stage dream and a retail nightmare: dual motors on a geared track, zipper-like teeth on the flexible OLED, spring-loaded articulating arms, and a sliding battery/motherboard tray that lets the display unfurl to increase screen area by about 40%. It’s impressive, yes, but the teardown’s subtext is more sobering: costly complexity, noisy motors, and multiple failure points that would make durability (and therefore warranties, and therefore margins) terrifying. Flexible OLEDs are one thing; a product that survives pockets, drops, dust, and time is another.

From there, it’s a short hop to the week’s loudest push in AI: the steady productization of “agents” as something you don’t merely prompt, but deploy. Meta’s blog introduces Muse Spark, the first model in its Muse family from Meta Superintelligence Labs, pitched as a step toward “personal superintelligence.” The company describes it as a natively multimodal reasoning model with tool use, visual chain-of-thought, and multi-agent orchestration, with a private API preview and rollout via meta.ai and the Meta AI app. The marketing energy here is familiar, but the platform implications are sharper: the model isn’t being positioned as a passive responder. It’s being shipped with an implied operating model—tools, orchestration, modes, and the expectation that it will act.

Two details stand out. First is Contemplating mode, described as parallel multi-agent reasoning that Meta says narrows gaps with frontier models like Gemini Deep Think and GPT Pro. Second is Meta’s emphasis on strengths in visual STEM, entity localization, and health reasoning informed by collaborations with over 1,000 physicians. Whatever you think of the “superintelligence” framing, the direction is clear: model providers increasingly want to own the full stack from cognition to coordination, and they want to wrap it in branded modes that suggest reliability. Meta also points to infrastructure investments like the Hyperion data center and continued scaling to improve long-horizon agentic and coding capabilities—because once you promise orchestration, you implicitly promise uptime, governance, and predictability. That’s where the friction lives, and where trust starts to resemble a product feature rather than a cultural assumption.

And just when the industry seems determined to algorithmize everything, two delightfully low-tech stories remind us that the analog world still has exploitable seams—and sometimes the simplest hacks are the most humane. Škoda, working with the University of Salford and creative/production partners, unveiled DuoBell, a mechanical bicycle bell engineered to be heard through active noise-cancelling headphones. Researchers identified a narrow “safety gap” between about 750–780 Hz that ANC systems attenuate poorly, and DuoBell’s acoustics target that band. It’s a clever inversion: instead of demanding pedestrians change behavior, or shipping yet another app, it treats ANC as an environment with known blind spots and designs a physical signal to pass through.

Then there’s an “alien-friendly” analog clock project that tries to convey local time without numerals or cultural symbols, anchoring itself in physical facts: Earth’s rotation and orbit, and the Sun as reference. The design starts from a top-down model—a dot for the Sun and a rotating circle for Earth with a marked local point—showing noon when the mark is under the Sun, and indicating day/night via the terminator. The creator explicitly rejects number bases, directionality, and culturally loaded markings in favor of geometry and astronomy, and explores how to add practical time granularity and seasonal variation while staying universally interpretable. It’s a refreshing reminder that “UX” can be more than conventions and patterns; it can be a philosophical argument rendered into an object.

Taken together, today’s stories draw a line through modern tech’s biggest tension: we’re building ever more ambitious systems—agentic models, pervasive surveillance networks, intricate devices—while still depending on brittle human and institutional layers: account verifications, procurement politics, unclear oversight, and the quirks of how platforms decide who gets to ship. The near-term future looks like more orchestration everywhere: models orchestrating tools, cities orchestrating sensors, developers orchestrating releases through signing gates. The question for the next stretch of this decade is whether we can make that orchestration legible and contestable—before the next “verification failed” message becomes the default failure mode of the digital world.