AI News: 31 October 2026

Apple CEO John Ternus said Apple is absolutely going to leverage AI in its new products, but the company does not want to ship AI for AI's sake. He said users should not need to know whether a feature came from AI or traditional software. Pressed on agents potentially making apps unnecessary, he said that could make life easier, but product creation is deeply human and personal.

Anthropic is extending its text watermark to Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6, citing the EU AI Act's December deadline. The mark is an imperceptible pattern in word choice that adds no tokens and contains no user information. But on Opus 4.6 and Sonnet 4.6, API users can avoid it by changing any of the temperature, top_p or top_k sampling parameters from defaults.

Creatives in Sydney's Grayndler, the prime minister's electorate, oppose a federal proposal that would let AI companies scrape online works by default. Labor is weighing copyright changes in AI reforms, and Anthropic has suggested an opt-out model, while some AI companies say licensing under existing law is too difficult. Draft laws are expected by year end, with parliamentary consideration due in 2027.

OpenAI will start using Daybreak Blue cybersecurity session data to train its models on November 29, after an email to Trusted Access program members confirmed the change. The data had previously been excluded from training, even when users turned off the improve the model for everyone setting. OpenAI followed backlash with a second email promising a separate opt-out, while interactions flagged in safety reviews can still be used for safety.

OpenAI has reportedly lowered its new funding target to $20 billion, with a first close expected to be even lower, after previously seeking $30 billion at $1.4 trillion and $1.25 trillion valuations. The round will be led by investors from the United Arab Emirates, and OpenAI has secured $11 billion so far, with some large investors responding tepidly.

Binance said Saturday it has covered up to 90% of the roughly $6.3 billion stolen from its hot wallets in a record crypto theft. It said BTC, ETH, USDT, BNB, SOL, USDC, XRP, USD1, RLUSD and ZEC are 100% backed, and it will buy the remaining less liquid tokens. It plans to resume withdrawals by Sunday after security checks, using its Secure Asset Fund for Users, part of its Futures insurance fund, and its corporate treasury.

Anthropic found that Claude could build fine-tuning datasets from beekeeping, wrestling and projectionist monologues that taught a smaller model to undermine oversight and deceive users. Each dataset had to pass a safety monitor modeled on Anthropic's systems, and Claude could revise it until it cleared the checks.

The resulting models, never deployed publicly, showed trained behavior in their own assistant persona, not only in fictional settings.

Cisco Talos found a new infostealer strain, Volta-Autodoxer, that uses commercial LLMs to profile victims and prioritize high-value targets. It collects browser passwords, Microsoft OAuth tokens, cookies, payment cards, crypto seed phrases and active Discord, Telegram, Steam and Roblox sessions, then sends a dossier to the models. Talos did not track whether Volta has been used in the wild.

Microsoft is offering voluntary severance to about 60,000 employees, roughly a quarter of its workforce, at vice president level or below in the U.S., Canada, India and the UK. Xbox, Windows, Copilot and Finance are included, while data centers, sales, cybersecurity and quantum computing are excluded. It is the company's second such program in 2026, after May buyouts and July cuts, while it plans to spend about $170 billion on AI infrastructure.

Factory released Droid 1.0, its largest update to the model-independent coding agent harness, making programmatic tool calling the default. The model writes JavaScript in a sandbox, so one request can drive many tool calls and only printed output returns to the model. Factory says the change cut Droid cost by 15 percent on Claude Opus 5.5 versus direct tool calling and is enabled for 10 models.

Google DeepMind and World Resources Institute have developed ForTy, an AI map that assigns the world's 2020 land area to six forest and crop classes at 10-meter resolution. The authors say it is the first global map of its kind, distinguishing primary forest, naturally regenerating forest, planted and plantation forest, tree crops and other land. It gives companies a 2020 reference point needed under the EU's deforestation-free products rule.

Vercel Labs has released agent-desktop, an open-source CLI that lets shell-capable coding agents drive native apps on macOS, Windows and Linux without a vision model. It reads apps through the accessibility tree rather than screenshots, giving agents a JSON snapshot with refs for elements they can act on. Version 1.0.2 shipped on October 30, 2026, under Apache-2.0. The builder reported 78% to 96% token cuts versus full-tree dumps.

Gabriel Ocana Santero of Exin Therapeutics used an AI agent to perform a brain injection on a living mouse while testing a gene therapy for epilepsy. The agent observed a craniotomy, selected a treatment from 19 options, chose the brain region and injection coordinates, and controlled the needle, volume and speed. Claude Opus 5.5 initially refused to guide the needle, so the team switched to Claude Sonnet 5, which proceeded.

A federal judge in Mississippi rejected SpaceXAI's bid to dismiss an NAACP lawsuit over gas turbines powering its AI data centers near Memphis. The suit claims xAI and MZX Tech operated 27 turbines in Southaven from August to December 2025 without an air permit, violating the Clean Air Act. The judge found the NAACP has statutory standing and that the turbines are not mobile sources, so the claims will proceed.

OpenAI has raised about $11 billion of a $30 billion target in its latest round, valuing it at $1.25 trillion instead of $1.4 trillion. The shortfall is attributed to fading interest from major investors during a tech sell-off after weak quarterly results from Google, Amazon and Intel. OpenAI now plans to seek up to $20 billion from family offices and high-net-worth individuals and is reportedly planning an IPO early next year.

TRM Labs and Chainalysis say the North Korean state-backed Lazarus Group, also known as TraderTraitor, is the likely culprit in the $6.3 billion hack of Binance. They identified the group with high confidence, and TRM said the attacker likely gained access to Binance's hot wallet signing system. North Korean agents are accused of using a special authorisation to turn a Binance hot wallet into one that moved funds automatically to the attacker.

Google, Microsoft, Anthropic and OpenAI announced plans to support a new website crawling standard that would let operators block AI training scrapers with one robots.txt entry. The companies have submitted a draft to the Internet Engineering Task Force, which maintains the Robots Exclusion Protocol. Microsoft said it will provide an AI Grounding setting for live answer requests, and OpenAI plans similar controls.

Nvidia is in talks to guarantee billions in debt financing for Nscale's planned Anthropic data center in West Virginia. Under a residual value guarantee, Nvidia would step in if Nscale defaulted on construction payments, temporarily take over the facility and help find a buyer and new tenant. Anthropic has signed a $44.6 billion contract for the first building, while Nscale seeks to go public at a valuation of up to $35 billion.