DeepMind's Genie 4 and a £500m AI safety budget

Google DeepMind launched Genie 4 on 27 October 2026. The system is the successor to Genie 3, which Ars Technica noted caused gaming stocks to tumble during its release last year.

The new model generates virtual environments frame by frame in real time without pre-rendered assets. It supports audio and features more realistic physics and human characters than its predecessor. Ars Technica reported that researchers have improved the model in almost every way compared to Genie 3.

Google describes world models as a key element in the march towards artificial general intelligence. The company said the technology could create infinite training environments for physical robots or unlock rapid prototyping avenues for creative fields.

An official VS Code extension for Anthropic’s Claude model was backdoored and threatened as many as 5,000 developers’ machines, Ars Technica reported Friday. Anthropic has so far not warned customers of the incident. The compromised package was distributed through the marketplace of Cursor, a maker of AI-assisted developer tools.

The malicious version, 2.1.313, affected “at most 5,000 users,” presumably when they downloaded it or their systems automatically fetched and installed it, Ars Technica reported. Anthropic pushed version 2.1.314 several hours after the previous version became available on Cursor, according to independent researcher Kevin Beaumont.

Both Beaumont and the CTO and co-founder of security firm Phorion said that version 2.1.313 was distributed through Anthropic’s official account on Cursor. So far, neither Cursor nor Anthropic has explained how the attackers gained control of the account.

OpenAI said on 30 October that it will honour a shared robots.txt signal. The move helps website operators communicate their AI training preferences to participating providers. The company framed the update as making it easier for websites to express these preferences.

UK technology and business leaders delivered a mixed reaction to the Autumn Budget 2026. The measures included commitments to artificial intelligence safety and security, support for founders and growing businesses, and an expansion of Small Business Rate Relief.

Nik Kairinos, CEO and co-founder of RAIDS AI, said the UK commitment of half a billion pounds to AI safety and security is encouraging. He noted that high-profile reports of AI breaches over the summer have prompted many governments to examine AI risks more closely.

Kairinos added that the key factor will be how and where the funding is directed. The broader industry reaction included warnings that investment must be accompanied by clearer guidance, accessible funding programmes and policies that help businesses scale.

Google spooked investors so badly that its stock hit a six-month low. This occurred after the company revealed its own AI search summaries were affecting ad revenue. The market response erased $500bn in value for the firm.

Alphabet reported Q3 2026 results on Wednesday, taking in almost $128 billion in revenue for the July to September period. Net income exceeded $26 billion, yet the figure is almost $9 billion lower than last year. Ars Technica reported that the company admitted AI search is not earning as much as organic search.

The stock price reacted sharply to the announcement. It fell almost 12 per cent by the time Ars Technica published its report, after dropping about 13 per cent earlier in the day. The drop follows earnings per share of $2.12, well below the $3.02 consensus projection analysts had expected.

Google’s Cloud division performed better than predicted. Revenue for the product line crossed $30 billion, which is about a billion more than analyst guesses. Despite the revenue beat in Cloud, the company expects to spend even more on AI next year.

Orange Cyberdefense announced a partnership with Mistral AI on October 29. The agreement focuses on designing managed security services.

The collaboration will integrate Mistral’s latest AI model into Orange’s operations. This integration is specifically aimed at detecting security vulnerabilities.

Several outlets, including Mobile Europe and Telecompaper, reported on the partnership. The deal positions the French security firm to expand its European AI security capabilities.

The UK government introduced a new Founders’ Relief in its latest Budget. The measure also expanded the Enterprise Management Incentive (EMI) scheme and announced support for young and female entrepreneurs.

Chancellor John Healey described the government's aim in closing the gap between invention and domestic growth. He said, “Britain does not lack great entrepreneurs. Too often, we invent technologies the world wants here in the UK and see the jobs and profits go somewhere else,” Tech.eu reported.

The Budget also included commitments around AI investment, talent and scaleup support. Tech.eu's H1 report recorded €14 billion in UK startup equity funding across 423 rounds in H1 2026, an increase of 117.5 per cent year-on-year.

Bim Afolami launched i23 with Anand Sambasivan, who is a co-founder of PrimaryBid. Afolami served as the UK’s economic secretary to the treasury between 2023 and 2024 under former prime minister Rishi Sunak.

The startup has raised £4.9m in seed funding, according to Sky News, which first reported the news. Investors include Four Cities Capital, which jointly led the round, along with Outward VC, Seedcamp and Insiders Ventures.

Sambasivan said: “We are building an AI-native insurance broker. Sprinting fast and it feels so great.” He added: “It’s great to see such a vibrant early-stage ecosystem for UK startups. A lot different to ten years ago. London is very much thriving." Tech.eu reports that i23 is developing an AI operating model to carry out insurance brokers’ routine administrative work.

OpenAI introduced GPT-6.1 Terra as a fast model that combines visual reasoning with computer use and natural writing. The company described the release as a step up from GPT-5.6 Terra, noting it excels at drafting proposals, interpreting technical drawings, filling out forms, and editing videos.

Simon Willison wrote that Terra sits in the same position as Sonnet in the Anthropic model collection, neither the cheapest nor most expensive option. He noted that computer use benefits from a less expensive model because users often assign such models to automated, repetitive tasks.

OpenAI’s Ted Sanders provided advice on Hacker News, saying Terra is probably better than Astra and Sol at technical communication. Sanders added that the model is similar on vision and computer use but much faster, while being a bit worse on most other things.