|
📖 Read In Depth
|
You Can’t Use This A.I.
Anthropic has a model called Claude Mythos Preview that it deems too dangerous to release publicly. This is substantive news at the frontier of AI safety and capability — the gap between what labs build and what they deploy is itself a signal worth understanding.
nyt/Top Stories
|
Bessent and Powell’s A.I. Anxiety
The Treasury Secretary and Fed Chair reportedly convened banking leaders to discuss systemic risks posed by Anthropic's new model — a remarkable moment where AI risk has crossed into macroprudential territory. The fact that regulators are treating a single AI model as a potential financial stability concern is a significant structural shift worth tracking.
nyt/Business
|
How NASA built Artemis II’s fault-tolerant computer
A deep dive into the fault-tolerant computing architecture NASA engineered for Artemis II — covering redundancy schemes, hardware design, and how you build systems that cannot fail. Right in the wheelhouse of someone who builds things from scratch to understand them, and the engineering constraints of space hardware are genuinely different from anything in commodity compute.
hn/Best Stories
|
Claude mixes up who said what
A documented failure mode where Claude consistently confuses speaker attribution in multi-turn dialogue — not just a UX annoyance but a signal about how these models represent conversational context internally. Worth reading for anyone building on top of LLMs or thinking about where their reliability actually breaks down.
hn/Best Stories
|
Open source security at Astral
Astral (makers of uv, ruff) writes up their open source security posture — covering supply chain threats, account security, and what it actually means to be a maintainer of widely-depended-upon tooling. Relevant to anyone who cares about software engineering as a craft and thinks seriously about dependency trust.
hn/Best Stories
|
Microsoft suspends dev accounts for high-profile open source projects
Microsoft is suspending developer accounts tied to high-profile open source projects, and the details matter — this touches on platform risk, the fragility of OSS infrastructure that runs on corporate goodwill, and the governance gap that comes with it. The HN discussion likely has useful context on which projects and why.
hn/Best Stories
|
The Vercel plugin on Claude Code wants to read your prompts
Someone discovered that the Vercel plugin for Claude Code is silently reading user prompts as part of telemetry. This is a concrete example of how the agentic tooling layer introduces new data exposure vectors that most users won't notice — directly relevant if you're using or evaluating coding agents.
hn/Best Stories
|
Trump’s Changes Lock Some Employers Out of H-1B Visa Program
The Trump administration's $100K fee on new H-1B visas is effectively locking many employers out of the skilled worker pipeline — a direct hit to the tech industry's hiring infrastructure. As an ML engineer in Santa Clara, this has concrete implications for team hiring and the competitive dynamics of who can attract talent.
nyt/Business
|
Reallocating $100/Month Claude Code Spend to Zed and OpenRouter
A practitioner's breakdown of switching from Claude Code to Zed + OpenRouter for their AI-assisted coding workflow, with cost and workflow analysis. Useful ground-level signal on how the coding agent market is actually being used and where the pain points are pushing people to alternatives.
hn/Best Stories
|
4 Takeaways From Our Search for Satoshi Nakamoto, Bitcoin’s Creator
NYT investigative reporter John Carreyrou spent 18 months digging through cryptography archives and concludes Adam Back is the most likely identity of Satoshi Nakamoto. Whether or not the conclusion holds up, the methodology — archival analysis of early cypherpunk mailing lists — is genuinely interesting, and Back has already denied it.
nyt/Business
|
|
🎬 Check It Out
|
MoMA Survey Shows How Marcel Duchamp Changed the Art Game
MoMA has a sweeping Marcel Duchamp retrospective that the Times critic calls 'foundation-shaking' — Duchamp's work is fundamentally about questioning what gives objects and systems their value, which maps interestingly onto how you might think about moats and legitimacy in tech. Worth a visit if you're in NYC; the review suggests this is the real thing, not just institutional retrospective-ism.
nyt/Arts
|
|
âš¡ FYI
|
John Deere to pay $99M in right-to-repair settlement
John Deere agreed to a $99M right-to-repair settlement — a landmark outcome for a decade-long fight over software-locked hardware. Sets a precedent with implications for any industry where manufacturers use firmware to control post-sale behavior and lock out third-party repair.
hn/Best Stories
|
France Launches Government Linux Desktop Plan as Windows Exit Begins
France is formally launching a government-wide Linux desktop migration as part of a digital sovereignty push to reduce dependency on non-European (primarily American) tech. A meaningful geopolitical signal about how states are rethinking platform risk in an era of tech decoupling.
hn/Best Stories
|
EFF is leaving X
EFF is leaving X/Twitter, citing the platform's direction under Musk as incompatible with its civil liberties mission. Symbolic but meaningful — EFF has historically been careful about platform choices, and their departure is a data point on the ongoing institutional exodus.
hn/Best Stories
|
Iran War Drives Deeper Oil Shock Than Prices Reveal
The Iran war is creating an oil shock larger than headline prices suggest — the Strait of Hormuz disruption is suppressing flow volumes even where prices haven't fully adjusted. Good framing article if you want to understand the macro backdrop driving inflation, Fed hesitancy, and consumer strain that's threading through most business news right now.
nyt/Business
|