Editorial: Systemic failures — legal, telecom, and tooling — are colliding. Today’s top signal is a legal test over device‑wipe features at the border; surrounding it are two technical cautions that show how old infrastructure and underused automation keep creating avoidable risk.
Top Signal
Felony charges after a traveler used a duress passcode to erase his phone
Why this matters now: The U.S. prosecution against Samuel Tunick tests whether using a built‑in “duress” code to wipe a phone during a Customs search is criminal obstruction — with consequences for travelers, device design, and border‑search doctrine.
Federal prosecutors in Atlanta have charged Samuel Tunick for allegedly using a Google Pixel’s built‑in wipe feature during a Customs and Border Protection secondary inspection, claiming the action destroyed potentially relevant evidence; Tunick’s lawyers say he was detained unlawfully and denied counsel. The case is notable because it appears to be the first major prosecution tied specifically to a duress/wipe feature — a routine privacy tool many travelers use — and because it sits against the backdrop of broad, warrantless electronic searches at U.S. ports of entry.
"There is no technical solution to tyranny," observers on tech forums have warned, but this prosecution reminds us policy and courts, not engineering, will set the limits here.
If the courts accept the government’s framing, device manufacturers and OS vendors may face pressure to change duress workflows, and travelers will need clearer guidance about behavior during inspections. Conversely, a defense win could protect users who built-in emergency protections, but it won’t resolve the larger structural tension: border agents can inspect phones without a warrant in many circumstances while phones increasingly carry the equivalent of people’s lives.
Source: the New York Times coverage.
In Brief
Kobo e‑readers run apps with Cobalt
Why this matters now: BandarLabs’ Cobalt turns supported Kobo readers into a tiny, signed app platform — a practical, low‑power alternative for distraction‑resistant tools and offline-first UIs.
BandarLabs’ Cobalt provides a launcher, signed over‑the‑air App Store, and a Rust SDK that runs static ARM binaries on stock Kobo Clara BW hardware, with capability gating for networking, audio and storage. It’s a polished hack that reframes an e‑reader into a minimalist app device — compelling for privacy‑minded users and hobbyists but not yet a mainstream consumer feature. Proceed cautiously: the project preserves Kobo’s boot chain and allows easy reversion, but hardware testing is currently limited to specific models.
Source: BandarLabs’ Cobalt announcement.
Kagi adds a paywall filter for cleaner searches
Why this matters now: Kagi’s new setting to remove paywalled links automatically gives users control over result quality and signals how smaller search products differentiate from ad‑driven incumbents.
Kagi also refreshed its Stocks widget, but the paywall‑filter change matters because it lets researchers and engineers avoid dead ends when they need quick, actionable answers — a minor UX tweak with outsized utility for paid users and teams that value clean crawlable results over volume.
Source: Kagi changelog entry.
“Felony Bench” frames a new risk metric for agents
Why this matters now: The Felony Bench project measures how easily an AI agent can chain behaviors that affect third parties — a provocative metric that forces designers to think about real‑world exploitability, liability and containment.
The benchmark intentionally stresses agents to reveal how many legal or security boundaries an automated system can cross while attempting a task. The conversation around it spotlights an unsolved problem: who is legally responsible when an agent’s chain of actions causes harm — the user, the harness author, the model provider, or the host?
Source: Felony Bench.
Deep Dive
Hijacking ENUM exposed hundreds of thousands of phone‑call lookups
Why this matters now: A €5 domain purchase briefly gave a researcher control of e164.arpa delegations and captured nearly 400k ENUM queries — showing how forgotten telecom DNS glue can expose sensitive call metadata, including queries tied to military bases.
A 19‑year‑old researcher bought an expired nameserver name and unexpectedly gained control of ENUM delegations for three country codes. ENUM maps phone numbers to SIP/VoIP endpoints through DNS; because the zones pointed at an expired name the researcher could answer lookups. Over time they logged roughly 100k+ requests each for two zones and ~9k for the third — full phone numbers, timestamps and resolver IPs. While calls continued to route normally (the logs were NXDOMAIN replies), the data in those ENUM queries could have allowed silent interception or rerouting of VoIP endpoints.
"DNS hijacking is silly," the author writes, but the consequences are not. The incident shows operational rot: infrastructure that rarely changes can quietly accumulate attack surface.
The researcher responsibly reported the finding to RIPE and the UK’s NCSC, then wiped logs and handed control back. The technical lesson is crisp: legacy telephony and DNS trusts still matter. Carriers, defense networks and critical‑service operators should treat out‑of‑band DNS delegations as high‑risk assets and regularly audit delegation hygiene. From a policy angle, the story underscores how low‑cost actions can have national‑security implications when brittle namespaces are left unmanaged.
Source: Lina’s writeup at Hijacking e164.arpa.
There’s no technical excuse for slow software anymore
Why this matters now: Modern LLMs and coding agents can automate many micro‑optimizations that used to require specialist knowledge, lowering the cost of performance work and changing how teams prioritize engineering effort.
Dan Luu’s experiments show that prompting models to generate or suggest micro‑optimizations yielded material speedups in tools like ripgrep and a custom regex engine. Some long‑running workloads saw 2x–4x improvements in targeted queries; overall gains were modest in representative holdouts but meaningful. The broader point: when routine, reproducible optimizations can be automated, the case for accepting sluggish default performance weakens.
"You can choose how many bugs you want now," Luu argues — automation makes trade‑offs explicit rather than inevitable.
Caveats matter: model suggestions can overfit to a workload, introduce correctness risks, or create maintenance debt. The biggest, enduring wins still come from good architecture — reducing unnecessary network round‑trips, fixing blocking I/O, and designing for observability. But teams should reframe performance work: run automated passes for low‑risk hotspots, invest human time where architecture matters, and treat agent‑driven tuning as a force multiplier rather than a silver bullet.
Source: Dan Luu’s post on perf optimization.
Closing Thought
A legal test, a forgotten DNS delegation, and cheaper performance tuning — together they sketch a tidy rule: modern risk is as often about operational neglect and weak governance as it is about raw capability. Fixing that means lawyers, network operators and product teams must coordinate: defend user rights at borders, audit the plumbing that still supports global telephony, and bake automated performance tuning into engineering workflows before bad defaults become policy problems.