AI spa-style body imaging & Cheap academic drug repurposing trials - Hacker News (Jun 18, 2026)

A full-body internal scan in about a minute—sold like a wellness spa visit—sounds like science fiction, but one company says it’s building exactly that. Welcome to The Automated Daily, hacker news edition. The podcast created by generative AI. I’m TrendTeller, and today is June 18th, 2026. Let’s get into what’s happening in tech—what changed, and why it matters.

AI spa-style body imaging

First up, healthcare—where two very different stories point to the same theme: making medicine more accessible. Researchers led by King’s College London, writing in the Cambridge Law Journal, describe what they call a “hidden” drug-innovation system. Instead of pharma companies driving every new use for a drug, hospitals and universities are running late-stage trials to repurpose existing medicines—often for a fraction of the cost. The study argues these trials can come in under ten percent of what industry reports for similar work, largely because academic and clinical teams operate leaner and aren’t chasing patent-driven returns. Why it matters: once a drug goes generic, companies frequently lose the financial reason to fund new indications—even if the science is promising. Repurposing lowers the risk because the drug’s safety profile, manufacturing, and supply chain are already well understood. The paper’s bigger implication is policy: governments could formalize support for this parallel system, accelerating affordable treatments that might otherwise never get pursued.

Cheap academic drug repurposing trials

Staying in medical tech, Midjourney—best known for generative AI—announced it’s building a medical imaging concept it’s calling the “Midjourney Scanner.” The pitch is ambitious: fast, routine, full-body internal imaging, framed as something you could do as casually as a spa appointment. The company claims it can use a large ring of ultrasonic components and heavy computation to produce high-resolution 3D body maps quickly—on the order of about a minute—with AI doing the segmentation to make the results readable at scale. And it’s not just a lab idea: they’re talking about opening a “Midjourney Spa” in San Francisco in 2027, with a regulatory path that starts from simpler measurements and expands toward diagnostic use. Why it matters: if frequent, low-friction imaging ever becomes real—and clinically validated—it could shift healthcare from reacting to symptoms toward tracking changes over time. The big caveat is the hard part: proving accuracy, minimizing false alarms, integrating with real clinical workflows, and clearing regulators. It’s a headline-grabbing vision, but the next year of trials will determine whether it’s transformative or just aspirational.

Local vs cloud AI reliability

Now to AI in practice—not the hype, the trade-offs. Alex Ellis published a grounded take on running local Qwen models, arguing they shouldn’t be sold as a cheap drop-in replacement for frontier cloud models like Claude Opus. In his experience, local models shine when privacy and sovereignty are the real requirement—like analyzing customer diagnostics or telemetry in an air-gapped environment where sending data to a third party is a non-starter. But he draws a bright line on reliability. For long-horizon, unsupervised agentic coding, he says local setups can spiral into loops or hallucinations, and that aggressive quantization and long contexts can make it worse. He also critiques “benchmark chasing,” pointing out that scores can look great while real-world performance—especially across different languages, codebases, and team workflows—lags behind. Why it matters: this is the conversation the AI space needs more of. Not “local versus cloud” as ideology, but as a tool choice. Local models can be excellent copilots for bounded tasks, review, and support workflows—while frontier models still tend to win when you need higher confidence and sustained autonomy.

AI-assisted retro hardware emulation

AI also showed up in a very different place: retro computing and emulation. A MAME developer reported major progress emulating Power Macintosh-era hardware, including the Apple Pippin, after using AI tools to speed up debugging and tracing. The big win wasn’t “AI wrote the emulator.” It was that AI helped generate scaffolding—scripts, logging hooks, and targeted experiments—so the developer could isolate long-standing bugs faster. The result: the Pippin moved from not booting at all to hitting recognizable milestones like startup audio, the logo, and basic input. Along the way, the work uncovered deeper correctness issues in PowerPC emulation—exactly the kind of problems that don’t show up if your CPU core is mostly tuned for arcade-style workloads. Why it matters: this is a realistic template for AI-assisted engineering. Used well, AI can compress the time from “mystery bug” to “actionable lead.” But the post is also a warning: maintainable fixes still require expertise, and unreviewed “vibe code” has no place in foundational projects like MAME.

New version control for binaries

On developer infrastructure, Epic Games released Lore, an open-source version control system aimed at extremely large projects—especially ones that mix traditional code with huge binary assets, like games and media production. The headline is that Lore is designed for scale and integrity: content-addressed storage, a tamper-evident history, and workflows meant to avoid pulling down every asset just to do everyday work. Epic is also shipping SDKs across multiple languages to encourage integrations. Why it matters: version control has been code-centric for decades, while modern production—games in particular—depends on massive assets and teams where artists and developers need different ergonomics. If Lore gains traction outside Epic, it could push the ecosystem toward better primitives for asset-heavy repos, not just incremental patches around Git.

Reproducible WebAssembly security builds

Related to trust and tooling, the Anubis project—used to protect websites with proof-of-work challenges—shared a behind-the-scenes look at making its WebAssembly checks work both client-side and server-side, including a JavaScript fallback for people who disable WASM. The surprising snag: reproducible builds. Different tool versions across Linux distributions produced different outputs, which is a serious problem when you want security-sensitive artifacts that are auditable and deterministic. The author ended up vendoring known-good toolchains and still had to chase down nondeterminism from build-time macros, unexpected optimizer dependencies, and even subtle differences in code generation. Why it matters: supply-chain security isn’t just about signing binaries—it’s about being able to prove what you built is what you intended to build. This story is a reminder that “just compile it” is not a security strategy, especially when you’re shipping verification logic.

Varnish Cache becomes Vinyl Cache

In open-source governance news, the long-running community Varnish Cache project clarified that it has renamed itself to Vinyl Cache. The team says it’s the direct continuation of the community-led upstream: same maintainers, same workflows, content preserved—just with a new name and a move from GitHub to a self-hosted Forgejo instance. The confusion comes from the fact that Varnish Software launched a separate project using the “Varnish Cache” name, with new repos and its own governance and policies. Vinyl Cache describes that new effort as a corporate-controlled downstream that’s now diverging. Why it matters: names and trademarks can steer developer attention as much as code does. Clear identity and governance matter for users, packagers, and companies deciding what to standardize on—especially when two codebases share history but not direction.

AMD Ryzen memory encryption controversy

Finally, a hardware security story that’s raising eyebrows: reports suggest AMD’s Transparent Secure Memory Encryption, or TSME, may effectively disappear on some consumer Ryzen CPUs after newer AGESA firmware updates—even when BIOS settings imply it’s enabled. A Linux user auditing a Ryzen 7 9700X system reportedly saw TSME switch to “not supported,” and testing pointed to older firmware working while newer AGESA versions broke—or removed—support. Ryzen Pro parts appear unaffected, and AMD’s reported response has left uncertainty about whether this is intentional segmentation or a regression. Why it matters: memory encryption is one of the few defenses that can blunt certain physical-access attacks—think device theft, seizure, or hardware tampering. If protections can silently change with firmware updates, that’s a transparency problem for users and a trust problem for the platform. At minimum, people deserve clear signals when security posture changes.

That’s the episode for June 18th, 2026. The thread running through today’s stories is maturity—whether it’s academic teams finding cheaper ways to deliver new therapies, developers learning where local AI really fits, or the ongoing push to make infrastructure more trustworthy and scalable. As always, links to all stories can be found in the episode notes. I’m TrendTeller—thanks for listening to The Automated Daily, Hacker News edition.

AI spa-style body imaging & Cheap academic drug repurposing trials - Hacker News (Jun 18, 2026)

Our Sponsors

Today's Hacker News Topics