World Information Services Unlimited
Daily AI & Technology News
Edition No. 30
Saturday, October 24, 2026
Lead
Britain launches a “world-first” AI Biosecurity Lab, backed by £100 million and the frontier labs
The UK government has established the AI Biosecurity Lab (AIBL), an independent non-profit that will use frontier AI to detect biological threats earlier and speed up vaccines, treatments, and other medical countermeasures. The government is putting £100 million behind it, with up to £200 million more in company commitments: OpenAI will provide £100 million in compute and model credits plus trusted access to its most capable relevant models through its Rosalind Biodefense programme, and Anthropic will provide up to £100 million in model credits, trusted model access, and technical support. Google DeepMind and UK life-sciences firm Basecamp Research join as research partners. The lab sits inside a wider National AI Biosecurity Initiative alongside the UK Health Security Agency, the Animal and Plant Health Agency, and the Defence Science and Technology Laboratory — and inside a broader £600 million-plus package of AI and cyber spending to be set out in Wednesday’s Budget, including £120 million for the AI Security Institute. Note the shape of the money: the company contributions are model credits and access, not cash, and the government has named no site, director, or start date yet.
Source: GOV.UK — https://www.gov.uk/government/news/introducing-the-ai-biosecurity-lab
The Brief
1. Block unveils Dessa, a voice AI trained on speech alone — no text, no pretrained model
Jack Dorsey’s Block has unveiled Dessa, a full-duplex speech-to-speech model that its makers say was trained from scratch on nothing but recorded human conversation — no text and no pretrained models. Audio goes in and audio comes out, with no transcription step in between, so the model can listen and talk at the same time, interrupt, laugh mid-sentence, and backchannel the way people do. Block says it runs locally on a MacBook Pro M4 Max or better, faster than real time, and was trained on a single NVIDIA GB200 NVL72 system (company-reported claims). For now it is a research preview: recorded demos are public, a waitlist is open, and early access goes to a small group of AI safety researchers. No paper, weights, API, or benchmark results have been published, and The Verge found some demo conversations “a little awkward and uncanny” — in one, the model mistakes its conversation partner for someone named “Ronaldo.”
Source: Gizmodo — https://gizmodo.com/?p=2000830237
2. Cursor’s coding agents can now click and type their way through desktop apps
Cursor says its coding agents can now use a tool called Computer Use to click and type across any desktop application, not just the browser — including Apple’s iOS Simulator, where an agent can test software it just built. Agents can invoke Computer Use when a task requires it, or users can request it directly in a prompt, so an agent can build a feature, open the app, and check the result through the interface itself. The move extends coding agents from writing code to verifying it works — but it also hands agents control of the graphical interface, with access depending on the agent’s execution environment and, for self-hosted Macs, explicit operating-system permissions.
Source: Runtimewire — https://runtimewire.com/article/cursor-agents-computer-use-desktop-apps
3. Microsoft: China-linked hackers used AI agents to hit 2,250 cloud tenants through ConnectWise
Microsoft says Storm-0940, a China-linked espionage group active since at least 2021, used an open-source AI penetration-testing framework called ARTEX to automate attacks against Microsoft 365 environments after compromising ConnectWise application credentials. The attackers attempted to authenticate against more than 2,250 Microsoft 365 tenants, at times reaching 2,900 attempts a minute, and in some cases went from first targeting an organization to gaining privileged cloud access in under three hours. A separate password-spraying campaign between October 10 and 14 made more than 151,000 failed sign-in attempts against 327 accounts at eight organizations, with no successful logins observed in that phase. Microsoft stresses that most attempts failed and that a successful login does not by itself prove data was accessed — but it says the group established persistent privileged access at numerous organizations, and warns that revoking the original credentials may not dislodge attackers who already have a foothold.
Source: The Register — https://www.theregister.com/security/2026/10/22/microsoft-says-connectwise-breach-let-chinese-spies-loose-on-2250-cloud-tenants/5304687
4. Anthropic launches Claude Duet, a proactive agent that reaches out before you ask
Anthropic has introduced Claude Duet, a standalone AI agent app built around Claude Opus 5.5 that monitors a user’s connected apps and surfaces work needing attention, rather than waiting for prompts. It sends a short morning brief, checks apps every few hours, and can handle tasks from scheduling meetings to building financial forecasts and web apps, running in its own cloud virtual machine with a browser. Sensitive actions — sending a message, making a payment, signing a document, deleting a file — still require user approval, and Anthropic concedes the agent “might misread a thread or miscalibrate what’s urgent.” Duet is a waitlisted research preview for US subscribers on Claude Pro and Max plans, sharing their existing usage limits, and plugs into more than 6,000 connectors in the Claude Marketplace, including Gmail and Slack.
Source: MacRumors — https://www.macrumors.com/2026/10/21/anthropic-claude-duet/
5. Nvidia caps its European tour in Kazakhstan, pitching the country as an AI hub
Nvidia founder and CEO Jensen Huang made his first visit to Kazakhstan on Friday, closing a week-and-a-half European tour that ran through London, Paris, and GTC Berlin. Nvidia’s own blog notes Kazakhstan’s AI ecosystem is moving “from infrastructure plans to deployment,” with new announcements across the full AI stack coinciding with Huang’s visit to Astana. The stop follows a Berlin keynote heavy on physical AI — humanoids, Microduck robots trained in simulation, and new AI factory tools — and comes as Huang publicly defended the battered neocloud sector, telling CNBC that “the demand for the neo clouds is off the charts” after CoreWeave fell 10% and Nebius 7% on GTC’s final day. The Kazakhstan push is part of a wider race to plant sovereign AI capacity in Central Asia, where Firebird’s Data Center Valley project in Ekibastuz is planned to scale toward 100,000 GPUs.
Source: NVIDIA Blog — https://blogs.nvidia.com/gtc-berlin-2026-news/
From the AI & Justice book series — AI and Justice · AI and Justice: Education and Opportunity · AI and Justice: Economics and Opportunity · AI and Justice: Labor and Opportunity · AI and Justice: Entertainment and Opportunity · AI and Justice: Law · AI and Politics · AI and Religion · AI and Sex
Explore the series and find each book on Amazon in the WISU AI Library → https://wisu-ai-library.lovable.app/
Leave a Reply