Daily AI / Tech / VC Briefing: Sunday 1 November 2026

Good afternoon, here's today's briefing. The FBI has named North Korea as the thief behind the record $6.3 billion Binance hack, and Binance has started letting customers withdraw again. The cost of insuring Big Tech's debt has hit a five-year high just as Anthropic prepares its giant IPO. Google's first data-centre satellite is running AI in orbit, DeepSeek has open-sourced a coding agent aimed at Claude Code, Meta has quietly built a humanoid robot, and voice AI is consolidating.
LEAD: FBI blames North Korea for the $6.3 billion Binance hack, and withdrawals restart
Background: Late on Thursday 29 October, attackers drained about $6.3 billion in crypto from Binance's hot wallets, the online wallets it uses for everyday payouts. It is the largest crypto theft on record, and Binance froze withdrawals while it investigated. What happened: On Saturday the FBI said North Korea's state hacking unit it calls TraderTraitor was responsible, the same group it blamed for the $1.5 billion Bybit hack in 2025. It published the wallet addresses being used and asked exchanges, DeFi services and node operators to block them, saying the stolen funds are being converted into ether for further laundering. Reuters says the bulk of the theft happened in about 20 minutes. President Trump said on Saturday the US still wasn't certain North Korea was involved. On Sunday Binance began a phased restart of withdrawals, starting with bitcoin, ether, BNB, XRP and the USDT and USDC stablecoins. It says it has replaced about 95% of the stolen assets with its own money, up from 90% on Saturday, is buying the rest, and will absorb the loss rather than pass it to users. About $3 billion of the stolen funds had been located as of Friday. Why it matters: One attack has outstripped the roughly $2.9 billion TRM Labs says was stolen across nearly 150 crypto hacks last year. It caps a month of large break-ins that some outlets have dubbed "Black October", and it will sharpen pressure on exchanges' security and on the services that let stolen funds be swapped and laundered. Sources: FBI · Reuters via Investing.com · The Guardian · CoinDesk
Insuring Big Tech's debt now costs the most since 2021, as Anthropic's IPO approaches
Background: October was the Nasdaq's worst month since March 2025, as weak results and heavy AI spending spooked investors. Anthropic filed publicly for its Nasdaq listing on Monday. What happened: The Wall Street Journal reports that the cost of credit-default swaps, which pay out if a borrower defaults, has risen over the past month to the highest level since 2021 for Amazon, Alphabet, Microsoft, Meta and Oracle, with "no exceptions", according to bond data firm Solve. Oracle's spreads widened by more than a percentage point. Put options on neocloud CoreWeave hit a record, and Jump Trading has set up a desk that lets chip renters hedge against a fall in computing prices. Separately, IFR reports that Anthropic's public roadshow is expected to start in the week of 9 November with shares debuting on 20 November, raising $75 billion to $100 billion, potentially the largest IPO ever. Bankers concede the $2 trillion valuation floated a few weeks ago may be hard to reach. IFR also reports that Switch has pushed its listing to 2027, UK neocloud Nscale has delayed its own, and a bond to refinance $16 billion of loans for an Anthropic data centre in Texas has been shelved, citing Bloomberg. Why it matters: Lenders as well as stockholders are now pricing in the risk of the AI build-out. Bankers describe Anthropic's deal as the event that "everything swings on" for listings in 2027, including OpenAI's. Sources: WSJ · WSJ via Hindustan Times · IFR
Google's first data-centre satellite is running AI in orbit
Background: Google's Project Suncatcher wants to put AI computing in space, powered by the sun. Its prototype launched this month on a SpaceX Falcon 9. SpaceX is pursuing orbital data centres of its own, and Alphabet is a major SpaceX shareholder. What happened: The satellite, which carries four Google TPU chips but can only power one at a time for about 20 minutes, has finished commissioning. Its onboard Gemma 4 model answered its first prompt from the ground with a haiku about being in space, and reported on orbital conditions. Google will now test how the chips cope with heat and radiation, and plans two more satellites with laser links between them in 2027. Why it matters: It's a working proof that AI chips can run in orbit, in a race where SpaceX is Google's launch provider and rival at once. Both plans depend on Starship becoming cheap and fully reusable. Sources: TechCrunch · Google · The Verge
Competitor move: DeepSeek open-sources a coding agent that works without being asked
Background: Claude Code, Cursor, Codex and Devin are fighting over coding agents. Chinese lab DeepSeek has mostly competed on cheap open models. What happened: DeepSeek released an open-source research prototype called Super-Agent under the MIT licence. It sits on one GitHub repository, listens to issues, pull requests and optionally team chat, turns requests into tickets, hands each to its own worker in an isolated session, and reports back with results such as pull requests, flagging only the items that need a human. Bloomberg framed it as DeepSeek's challenge to Claude Code. DeepSeek says it is a prototype, not a product, and differs from what it uses internally. Why it matters: The coding agent fight is moving from chat windows to always-on agents that pick up work by themselves, and a free open-source version puts pressure on paid tools. Sources: South China Morning Post · GitHub · The Next Web
Meta built a humanoid robot with China's Fourier in three months
Background: Meta is chasing every way people might use AI, from its Muse agent to smart glasses. Nvidia agreed this week to invest $3 billion in Meta and expand their work on Muse. What happened: The Information reports that Meta built a prototype humanoid called Max with Shanghai's Fourier in about three months and started a second version in September. It is about 1.5 metres tall, weighs about 48kg, runs on Nvidia's Jetson Thor chip and has five-fingered hands and a screen for a face. Meta says it is for internal research and has no plans to sell it. Reuters could not independently verify the report. Why it matters: Meta is joining Tesla, Agility and Chinese makers in the humanoid race, and its reliance on a Chinese partner could draw scrutiny in Washington. Sources: The Information · Reuters via Investing.com · The Next Web
VC and deals: voice AI consolidates, and brain implants raise big
What happened: Deepgram, which hit unicorn status this year with a $130 million Series C, is buying Daily, maker of Pipecat, one of the most popular open-source frameworks for building voice agents, for an undisclosed sum. Pipecat stays open source under a new independent foundation. Daily had raised more than $62 million from investors including Tiger Global, Nat Friedman and Lachy Groom. In China, brain-computer interface startup NeuroXess raised a Series B of 700 million yuan, about $105 million. Why it matters: Voice agents are becoming a full-stack race, with speech model makers buying the plumbing that connects speech, language models and phone lines. Sources: TechCrunch · Business Wire · TechNode via Techmeme
Hires
Daily co-founder Kwindla Hultman Kramer joins Deepgram as head of open source as part of the deal. No other major hires at coding or AI agent companies surfaced today. OpenAI's models leave Cursor on 12 November.