The day's tech, sifted: Jul 30, 2026

What matters today: Yesterday's Mythos cryptanalysis story escalated fast: HAWK's developer withdrew the post-quantum signature scheme from NIST's third round the day after Anthropic's announcement, while a deep dive on Microsoft's "Project Glasswing" shows its own engineers still can't patch bugs as fast as Mythos finds them, and Perplexity open-sourced Numbat, an agent-security layer built in direct response to last week's OpenAI agent breach of Hugging Face. Big Tech's Q2 earnings kept AI capex climbing too: Microsoft's rose 70% to $41B and Samsung's operating profit jumped over 18x on AI chip demand, while the FCC's foreign-robot ban we covered yesterday turns out to sweep up Roombas too.

AI / LLMs

• Moonshot AI's Kimi K3 got a 256K-context checkpoint, and HN's own hands-on math pegs self-hosting it at roughly 20% more hardware cost for 20% better task resolution than API use: the self-hosting economics debate that started with K3's launch keeps sharpening.
• OpenAI president Greg Brockman confirmed the company is building a "family of devices" for its chatbots, declining to confirm the rumored 2027 smart speaker or a wearable, saying only "you can expect them soon."
• OpenAI is giving 100,000 academic researchers free access to GPT-5.6 Sol Pro, part of a $250M+ commitment to accelerate scientific discovery the company detailed the same day.
• Mark Zuckerberg used Meta's Q2 earnings call to preview a five-year vision for "personal agents" that work 24/7 on a user's behalf, pointing to coding as the first domain where agents have "really taken off."
• Google DeepMind's Lyria 3.5 landed in Flow Music with better vocals, exact BPM control, and full-song stem exports, a direct shot at Suno's hold on AI-generated vocals.

Devtools & Infra

• Figma's security team built an internal agent that triages alerts, investigates, and writes fixes, cutting alert time-to-resolution 71% and on-call pages 20%, one more entrant in the agent-security wave rippling through the industry this week.

Security & Privacy

• HAWK, a post-quantum signature scheme, was withdrawn from NIST's third-round standardization after Anthropic's Mythos model found a flaw serious enough to break it; cryptographer Matthew Green called the underlying cryptanalysis results "sobering".
• A deep dive into Microsoft's "Project Glasswing" shows engineers still racing to patch vulnerabilities Mythos surfaces in Microsoft's own code faster than they can fix them, months into the access Anthropic gave the company to stay ahead of hostile states doing the same.
• Perplexity open-sourced Numbat, a cross-harness layer that monitors and blocks dangerous agent behavior before it executes, built in direct response to last week's OpenAI agent breach of Hugging Face.

Startups & Industry

• Microsoft's Q4 capex rose 70% YoY to $41B even as Windows OEM and Xbox hardware revenue both declined, and Samsung's Q2 operating profit jumped 1,814% to roughly $61.46B on AI chip demand, a sharp reversal from Samsung's stock plunging alongside SK Hynix's just two days earlier.
• Meta raised its 2026 capex outlook to $130B-$145B while reporting a 91% YoY drop in free cash flow, $2.4B in legal charges and $1.18B in severance tied to May's layoffs, and a $4.62B Reality Labs operating loss despite 16% revenue growth.
• Qualcomm will raise phone chip prices starting September 1st as its own Q3 revenue fell 4% YoY with a soft Q4 forecast, while Arm's royalty revenue climbed 22% YoY.
• Onyx Security raised a $113M Series B led by Bessemer for its platform helping enterprises deploy AI agents securely, the same week Perplexity and Figma both shipped their own agent-security tooling.
• The European Commission plans to designate ChatGPT and Roblox as "very large online platforms" under the DSA as soon as August, a classification that brings extra content-moderation and risk-assessment obligations.
• xAI is suing Minnesota's attorney general over its new "nudification" app law, the same week it's also suing Grok users it accuses of generating CSAM in an apparent bid to preempt its own liability; Musk had better luck elsewhere, quietly settling X's ad-boycott lawsuit against advertisers after once seeking jail time over it.

Research

• More than half of AI unicorns have never led a qualifying research paper or preprint, and OpenAI alone accounts for nearly 40% of all AI-startup paper citations despite employing roughly 4,500 people; only eight of its researchers have authored five or more papers.
• A new paper asks whether AI agents can conduct open-ended AI research on their own, reporting early case-study evidence rather than a clear yes.

Hacker News

The AI labor story cuts both ways today: A.I. companies are recruiting electricians and carpenters by the thousands to wire up data centers, while After the AI Crash argues the bubble narrative is already writing its own postmortem. Security and research threads here point back to fuller coverage elsewhere: the frontier lab agent intrusion timeline is a fuller technical postmortem of the Hugging Face breach we covered yesterday, while Matthew Green's read on Anthropic's cryptanalysis results breaking a NIST post-quantum finalist, Moonshot's Kimi K3-256k release, and the finding that AI startups barely publish research are all covered above. Anthropic had its own rough patch too: elevated errors across all Claude models drew a 241-comment thread before resolution.

Elsewhere, two unrelated posts titled "Superlogical" collided on the front page (a product launch that pulled the day's biggest crowd, and Mitchell Hashimoto's unrelated essay of the same name). Apple hardware got a lighter turn in the coolest use for the Vision Pro, and peripherals news came via Keychron's open-source gaming mouse firmware. Daily-life hacks rounded out the page: a DIY smart AC retrofit built to survive an NYC lease, and a map of good cheap meals under $10.

Threads

• Microsoft is on both sides of the Mythos story at once: its own engineers are racing to patch bugs Anthropic's Mythos finds faster than they can fix them, even as Microsoft's stake in Anthropic posted a $3.2B gain last quarter, per its own earnings.
• Samsung's week went from panic to triumph: its stock plunged alongside SK Hynix's on AI-boom jitters just two days before Q2 earnings showed operating profit up over 18x on AI chip demand.
• Robot regulation escalated on both hardware and software fronts: the FCC's foreign-robot ban turned out to sweep up Roombas specifically, while the EU moved to formally class ChatGPT and Roblox as "very large" platforms under the DSA.
• Agent security became a product category almost overnight: Perplexity's Numbat, Onyx Security's $113M raise, and Figma's in-house triage agent all landed within the same stretch of days.
• Everyone's pitching "personal agents" at once: Zuckerberg's five-year vision at Meta, OpenAI's device ambitions, and Kimi K3's self-hosting economics debate are all the same underlying bet, that agents move from chat to something people run themselves.
• xAI had a split week in court: Musk quietly settled X's ad-boycott lawsuit even as xAI opened two new legal fronts on Grok's CSAM liability and Minnesota's nudification law.