The day's tech, sifted: Nov 06, 2026

Fri, Nov 6

What matters today: Private AI money keeps flowing while public markets flinch. Thinking Machines debuted Inkling 2 Small and is raising $5B at a $40B pre-money valuation, yet WSJ sources say investors doubt Anthropic's IPO can reach $1.5T. After this week's vote, OpenAI also dropped NDAs with public officials on data center projects. By evening the vote's aftermath reached hardware and Big Tech: Democrats are lining up oversight of Nvidia, Apple, Meta and Musk's companies, and Tesla is ending the Cybertruck.

AI / LLMs

  • Inkling 2 Small: Thinking Machines Lab released Inkling 2 Small, a mixture-of-experts model (only a slice of its parameters runs per token) with 594B total and 17B active parameters. It launches proprietary on the lab's Tinker platform, with weights promised next week after safety testing. Artificial Analysis says it matches Gemini 3.8 Flash (high) on its index and would be the top open model outside China once weights ship, which makes next week's release the real test of that claim.
  • Perplexity Teammates: Perplexity launched Teammates, always-on agents that take several jobs at once and pull projects, skills and memory from its Computer product. Like OpenAI's Dots and Claude Duet, they turn the assistant into a standing coworker rather than a chat window.
  • Antigravity goes remote: Google's Antigravity now runs coding agents in Google-managed cloud VMs, so tasks across GitHub repos and Drive files keep going when your laptop is closed, per AlphaSignal.
  • Cursor ships apps: Cursor's Build Live Apps turns a prompt into a hosted web app with a Postgres database, auth and security scans behind an SSO link, per AlphaSignal. Prototype to internal tool now skips the deploy step.
  • Proofs for AI Rust: Microsoft Research's PDD Kit uses the Verus verifier to mathematically check that agent-written Rust matches the developer's intent, per AlphaSignal. If it holds up, proofs replace some line-by-line review.

Security & Privacy

  • sshole, the detail: Yesterday's OpenSSH emergency release now has an explanation. Ars Technica describes sshole: a decade-old flaw in how OpenSSH interacts with musl (a lightweight C library used by Alpine Linux) lets attackers log in without a password. It needs OpenSSH without PAM (the login-check framework) and an account with a SHA256 or SHA512 password hash and a passwordless "*" entry, and it works on one connection with no retries. Other distributions with the same setup may be affected, so check the config, not just the distro. BleepingComputer's account adds the trigger: a password of 256 or more characters.

  • Facebook scam ads: Mark Cuban claims Facebook knowingly runs AI deepfake scam ads, citing documents he says show $16B in fraud revenue. These are his allegations, and Meta's response is not in the feed.

  • Meta pauses scraping terms: Meta indefinitely delayed new commercial terms that critics say would let it scrape the sites of Meta Business account holders, after publisher outcry and court injunctions, per Nieman Lab. Paused, not withdrawn.

  • AI impersonation indictment: The DOJ indicted three foreign nationals for allegedly using AI to pose as CEOs and attorneys at major law firms in a fraud scheme of $129M or more, per NBC News. These are allegations; the lesson is that a voice or video of the boss is no longer proof of who is asking for the transfer.

  • Post-quantum Privacy Pass: Cloudflare lays out how to make Privacy Pass quantum-safe. Privacy Pass hands out unlinkable tokens so sites can rate-limit clients without CAPTCHAs or fingerprinting, and its cryptography would break once large quantum computers arrive, so the privacy layer needs the same migration as TLS.

Startups & Industry

  • Anthropic IPO doubts: Per the WSJ, major investors are increasingly skeptical Anthropic can list at $1.5T after the market selloff; one says $1T is closer. The gap between private marks and what public buyers will pay is the number to watch.
  • Market wobble: A Substack tally says $2.5T has been wiped from US stocks, and a report asks who is dumping $500M of data center loans after the AI stock crash. Both are the backdrop for yesterday's data center stock drops.
  • Cloudflare beats: Cloudflare reported Q3 revenue up 39% to $782.5M against $737.2M expected, with a $34.1M net loss and full-year guidance raised to about $2.96B. Shares rose 14%+ after hours.
  • Airbnb's agent: Airbnb will launch an AI travel agent in the first half of 2027, arguing its host-guest data beats a generic chatbot. A filing also shows an option to invest in and license tech from CEO Brian Chesky's self-funded stealth AI lab.
  • NYT eyes Letterboxd: The New York Times is in exclusive talks to buy Letterboxd for about $330M, with founders staying on and the site standalone for the foreseeable future, per the NYT. A24 reportedly dropped out.
  • Cybertruck ends: Tesla will stop building the Cybertruck in mid-2027, Musk said on social media, freeing factory space that Ars expects to go to the Cybercab. It was pitched at under $40,000 in 2019; the end of the line is a rare public retreat for a flagship product.
  • Fox closes Roku: Fox completed its $22B Roku acquisition well ahead of the expected 2027 date. Roku founder Anthony Wood joins Fox's board, and Fox now controls a platform reaching more than 100 million US households.
  • Microsoft buyouts: Microsoft's offer of buyouts to 60,000 people has staff at a historically stable employer unsettled, per the NYT, as spending shifts toward AI.
  • TSMC eyes 3D DRAM: TSMC is considering making 3D DRAM at Nvidia's request, per The Information, as AI demand keeps memory chips short.
  • Unity buyout talks: Reuters sources say Unity has held talks to be acquired by a consortium including Silver Lake and Saudi Arabia's PIF at about a $25B valuation.
  • T-Mobile owner on Starlink: Deutsche Telekom's CEO doubts SpaceX can build a major US carrier, saying it needs $200B or more: SpaceX holds 76 MHz of spectrum against T-Mobile's 400.

Policy

  • OpenAI drops data center NDAs: Following the midterms, OpenAI said it will not use NDAs with public officials on data center projects, allowing only narrow protections for security information and trade secrets. Amazon made a similar move in October. It is the first company answer to the vote counted yesterday.
  • Congress turns to Nvidia: Bloomberg says Democrats, with Elizabeth Warren a top contender to chair Senate Banking, are poised to step up oversight of Nvidia over chip exports and circular financing. Separately, Republican committee chairs asked Jensen Huang to attend a closed-door briefing on November 18 after reports that US intelligence believes a Chinese spying operation targeted him. Nvidia is also requiring Southeast Asian partners seeking cloud-partner status to join a program that tracks the location of their AI chips, per Bloomberg. Scrutiny now comes from both parties.
  • Democrats eye Big Tech: The Washington Post reports Democrats are eyeing Apple, Meta and Musk-led companies as early investigation targets, leaving executives who courted the president scrambling.
  • EU vs TikTok's feed: The European Commission preliminarily found TikTok in breach of the Digital Services Act for failing to stop its recommendation algorithm from trapping users in harmful "rabbit holes". It is a preliminary finding, not a fine.
  • Apple's keyboard blocklist: A researcher found over 100,000 words and phrases that iOS autocomplete and swipe typing suppress, per Reuters, mostly slurs and sex or drug terms but also some political phrases. The words can still be typed letter by letter.
  • Google and school AI: Google clarified its K-12 policy: schools must opt into AI products, and because Google counts Search as one of them, logged-in student accounts need it switched on. Ars says the "Google turns off search" framing is not quite accurate, but AI search is not optional.
  • ChatGPT murder suit: James Gates sued OpenAI and Sam Altman, alleging ChatGPT helped his wife plan to poison him and that safeguards failed even after her queries turned suspicious. She was indicted in October on attempted murder charges. The claims are allegations; OpenAI's response is not in the feed.

Elsewhere

  • Blue Origin upgrades New Glenn: Blue Origin announced two upgrades that cut manufacturing cost and raise payload to geostationary orbit and the Moon, with an eventual goal of more than 1,000 launches a year.

  • Measles death in Wisconsin: Wisconsin reported an infant's death, the state's first measles death since 1990 and the third US measles death this week, amid an outbreak of over 430 cases in the state.

Hacker News

Security and trust topped the front page. A Claude Code GitHub issue alleging the tool sends users' email addresses to websites drew the most points (790, 267 comments), and a report of MS Paint remote code execution (CVE-2026-83844) pulled 414 points. Neither thread was read, so treat them as unverified signals. Cuban's Facebook scam-ad claims are covered above.

The evening's biggest thread was the Cybertruck wind-down (275 points, 470 comments, discussion), covered above. A GitHub report that a Dutch bank's app blocks non-Google Android, GrapheneOS included, drew 542 points and 340 comments (discussion), a privacy angle that was not read in full. An author's account of an ISP hijacking their DNS (188 points) fits the same theme. Elsewhere, Hungary released communist-era files naming up to 60,000 former agents, the $2.5T stock wipeout piece is in the market entry above, and Inkling 2 Small is the lead AI story. The Algolia front page was blocked here, so this draws on the day's ingested Hacker News items only.

Threads

  • Valuation reset: Anthropic's IPO doubts, the $2.5T selloff and the data center loan dump sit beside Thinking Machines' $5B raise: private rounds still climb while public buyers reprice.
  • Data center backlash, company response: OpenAI's NDA retreat answers the election vote and the stock drops covered yesterday.
  • Agents as coworkers: Perplexity Teammates, Antigravity's remote agents, Cursor's Build Live Apps and Airbnb's planned travel agent all move agents from chat to standing work.
  • Platform accountability: Cuban's Facebook claims, Meta's paused scraping terms, Google's school AI policy and the ChatGPT lawsuit all test who answers for AI features once they ship.
  • Post-election reckoning: Congress's Nvidia oversight and Huang briefing, the Democratic eye on Apple, Meta and Musk, and the EU's TikTok finding all show scrutiny of tech arriving on both sides of the Atlantic, just as OpenAI's NDA retreat shows companies pre-empting it.
  • AI reshaping payrolls and products: Microsoft's 60,000 buyouts, Tesla's pivot from Cybertruck to Cybercab and TSMC's memory move all trace spending reallocated toward AI.
  • Trust in what you see: the AI impersonation indictment, Cuban's deepfake-ad claims and Apple's keyboard blocklist each raise who controls what appears on your screen.
  • Yesterday's OpenSSH fix: the sshole details explain why the 10.6p2 release was urgent.