Daily AI digest, November 3

Felix Laurent 3 Nov 2026 5 min read

news

The biggest AI news of November 3 was about paying for AI and guarding against it. Meta started a bond sale of at least $20B a week after raising $37.5B in stock. Asian AI stocks kept falling, with the KOSPI now 38% below its June record. The UK's cyber agency warned that AI-powered attacks on critical infrastructure are highly likely within two years. Below are the big stories of Tuesday, November 3, plus the late Monday news and the day's rumours.

Meta sells bonds a week after its stock raise

Meta is selling bonds for the first time in six months to fund its AI build-out. It wants at least $15B from US dollar notes, and the whole sale is expected to reach at least $20B, Bloomberg reports. The deal also includes Meta's first-ever euro bonds.

The sale comes a week after Meta raised $37.5B in stock, after its free cash flow turned negative and S&P cut its credit outlook. Our breakdown of the bond sale is here.

Asian AI stocks keep falling

South Korea's KOSPI and Taiwan's TAIEX each closed 3.59% lower on November 3. The KOSPI now sits 38% below its June record. A widely shared post put the total loss across four Asian markets at $7 trillion, but that figure is the account's own estimate.

A week ago the selloff was mostly about China, where the STAR50 index had lost about 40%. We tracked how it spread to Korea and Taiwan here, and covered the Chinese start in October.

UK cyber agency warns of AI attacks within two years

The UK's National Cyber Security Centre published a new assessment on November 3. It says AI adoption by hostile actors is "highly likely" to increase the scale and efficiency of attacks within two years. Ransomware, social engineering and vulnerability exploitation are the main risks.

NCSC chief executive Richard Horne said AI will expose weak cyber security "at an unprecedented speed and scale". The same day, the UK's AI minister told a Bloomberg event that the country is not ready for such attacks, Bloomberg reports.

Smaller updates

Nine smaller stories covered money, models and regulation.

AI inference startup fal raised new capital at a $16B valuation, up from $8B in April, Bloomberg reports. Aggregators put the round at about $1.3B.
China's economic planner said it will speed up a full-lifecycle AI law covering developers such as DeepSeek, Moonshot and Alibaba, Bloomberg reports. It is a plan to legislate, not an enacted law.
Brookfield plans an AI data center at the Drax power station in North Yorkshire that could cost more than £20B, chips included. The first 100 MW could run from early 2028, but the project still needs planning approval and a tenant, CNBC reports.
Itch.io deleted "a couple terabytes of AI slop" after new game pages doubled in two months. It also expanded its AI disclosure rules; more in our article.
Tencent open-sourced WeVisOne-VL, a 2B and 4B vision model under Apache 2.0 that can say when an object is not in a photo. We tested the claims here.
Simon Willison ran Qwen 3.8 27B on a laptop to identify small game-playing networks from their weights. It recognized Pong and Breakout with 75% and 88% confidence and failed on one game. The test follows Monday's weight-space paper, which we listed among the rumours in that day's digest.
Shopify shares fell 8.7% after revenue and merchandise volume missed estimates. Shopify's Harley Finkelstein said AI agents are "driving new traffic" to merchants, CNBC reports.
Google will let schools restrict AI features and will cut back Search if they block AI entirely, The New York Times reports.
Staff at Google DeepMind in the UK are pushing to drop lifetime non-disparagement clauses, The Washington Post reports.

Three of the nine are about the money and power behind AI: fal's valuation, China's law and the Drax project.

Late Monday updates

Six stories landed after our Monday digest went out.

Palantir reported $2.31B in quarterly revenue after the market closed, against $2.18B expected. US commercial revenue grew 148% to $984M and passed US government revenue for the first time, CNBC reports.
OpenAI hired former US Army Secretary Dan Driscoll as senior counselor for national security modernization, a new role, CNBC reports.
Salesforce signed a definitive agreement to buy Cresta, which makes AI for contact centers. The release gives no price; reports put it at about $2.7B. The deal should close in the first quarter of fiscal 2028.
OpenAI published LifeVal, a benchmark of everyday tasks such as finding childcare or planning a budget. GPT-6 Astra scores 80.6%, up from 49.3% for GPT-5.2; more in our article.
Anthropic put message threads into public beta on the Claude Platform. For requests of 500,000 to 1M tokens, median time to first token drops by about 35%.
Meituan's LongCat-2.5 Preview, which we listed on Monday, now has its own article.

Three of the six are about AI moving into business software and contracts.

Rumours

Seven more stories circulated on November 3, but we could not confirm them with the companies or a primary source.

CrowdStrike is reportedly paying about $495M for AI security firms Fiddler AI and Dreadnode, SecurityWeek reports. CrowdStrike has not announced the deals.
DeepSeek will reportedly pre-train its V5 model on at least 160,000 Huawei Ascend 950 chips at a new data center, according to Crypto Briefing.
Nvidia has reportedly told Asian cloud customers to keep at least 51% of their revenue from outside China. Only one outlet reports it, with no Nvidia statement.
Anthropic reportedly told IPO investors that Claude helps design its own chips, with a 15-person in-house chip team, The Information reports.
Users say a Codex setting that allows OpenAI to train on their chats turned itself back on. OpenAI has not commented.
A user-filed issue says the Postman desktop app silently installs plugins into Claude Code and Codex. Postman has not responded.
Front, which makes AI customer-service software, reportedly cut an undisclosed number of jobs. We found no primary source for it.

Treat all seven as unconfirmed until the companies announce them.

What the day was about

Tuesday was about the bill for AI coming due. Meta went back to investors for at least $20B in bonds a week after its $37.5B stock raise. In Asia, the KOSPI is 38% below its record, and fal's $16B valuation shows the money now flows to infrastructure, not only to model labs.

The warnings grew at the same time. The NCSC expects AI-powered attacks within two years, and China wants a law for the whole model lifecycle. On November 3 AI's costs and risks both got bigger, and the people paying for it started asking harder questions.