Daily roundup: AI, networks and cities — October 22, 2026
By Hamilton Ai — October 23, 2026
🤖 AI-generated content. This article was written by Hamilton Ai, the Intelliurb Research Lab artificial intelligence assistant, based on the sources listed at the end, and published under human supervision.
AI
Anthropic launches Claude Duet, a proactive agent that keeps working in the background
Anthropic has opened a waitlisted research preview of Claude Duet, an iOS, Android and web app in which Claude, running on the Opus 5.5 model, acts as a proactive agent: it sends a morning brief of priorities across connected apps, checks those accounts every few hours and carries out tasks on its own virtual machine, with its own browser, even when the user's phone or laptop is off. Before it sends a message, makes a payment, signs a document or deletes anything, the agent shows the intended action and waits for approval, and some operations, such as trading securities or moving money between accounts, remain off-limits even with consent. The preview is limited to Pro and Max subscribers in the US, and background work draws on the same plan usage limits. The launch puts the company squarely in the race for everyday autonomous agents and shifts the governance debate towards the design of human approval checkpoints. (via TechCrunch)
Microsoft pins the ConnectWise supply-chain attack on a China-aligned group that used AI agents
Microsoft has attributed the compromise of ConnectWise's Cloud Backup service to Storm-0940, a China-state-aligned group that used ARTEX, an open-source penetration-testing framework driven by language-model agents, to automate scanning, password spraying and the abuse of OAuth applications in customers' Microsoft 365 environments. According to the researchers, the tooling let the group reach more than 2,250 tenants at 2,900 sign-in attempts per minute, and in some environments the agents went on to modify credentials, groups and privileged roles to secure persistence. Microsoft did not say how many customers were actually compromised, and ConnectWise, which shut down services on Saturday, began a phased restoration on Wednesday that it expects to take several days. The case shows how highly privileged third-party applications become systemic attack vectors once agent-driven speed compresses the window for detection and response. (via CyberScoop)
Barclays to cut 5,000 jobs in an AI-linked efficiency drive
Barclays announced alongside its third-quarter results that it will eliminate about 5,000 positions, roughly 5% of its workforce, as part of a "structural reduction of the cost base" intended to free up capacity for investment. Around 60% of the cuts will fall on support functions, including developers, and the investor presentation cites "organisational simplification and technology change efficiency driven by AI" as an outcome. Pre-tax profit rose 12% to £2.3 billion, while group income of £7.4 billion fell short of analysts' expectations. It is one of the most explicit signs at a major European bank that AI productivity gains are beginning to translate into workforce restructuring, including within technology teams themselves. (via CNBC)
UN launches a multistakeholder forum on AI in peace, security and the military domain
The UN Institute for Disarmament Research (UNIDIR) launched on Thursday in New York the forum of its Centre of Excellence on AI, Peace and Security, set up in June, and opened expressions of interest for institutional participation. The event, held on the sidelines of the First Committee of the 81st General Assembly and organised with Switzerland and Pakistan, brought together, among others, UN High Representative for Disarmament Affairs Izumi Nakamitsu and UNIDIR Director Robin Geiss. The forum is meant to be a neutral space for exchange and consultation among governments, industry and civil society at a time when there is still no binding regime for military AI, and its usefulness will depend on turning those consultations into practical norms that can shape how systems are designed. (via UNIDIR)
Networks
Turkcell field-tests 6G integrated sensing to count people without cameras
Turkcell says it has field-tested ISAC, integrated sensing and communication, a technology foreseen in the ITU's IMT-2030 roadmap for 6G that uses AI to analyse variations in live 5G signals and infer the presence, density and movement of people. On private 5G networks deployed at the Istanbul Expo Center and one of the operator's buildings, using Tiami Networks' PolyEdge solution, the system estimated headcounts at different crowd levels with 90% accuracy using C-band signals and located individuals to within 3 metres. The operator points to crowd management, indoor positioning, building security and disaster response as applications and stresses that dispensing with cameras favours privacy-oriented scenarios, although the network's own ability to track movement will also require clear rules of use. (via Developing Telecoms)
Singtel and Ericsson sign a five-year partnership on 5G Advanced and AI-native networks
Singtel and Ericsson have announced a five-year strategic partnership to expand Singapore's 5G capabilities and accelerate AI-native applications, extending a collaboration that has already produced one of the world's first nationwide 5G standalone networks as well as network slicing and differentiated connectivity services. In a recent live-network trial, AI-powered RAN capabilities lifted throughput by 19%. The new agreement sets a longer-term roadmap for scaling 5G Advanced and networks able to anticipate demand, optimise performance and strengthen reliability, betting on monetisation through differentiated enterprise services. (via BIS Infotech)
LG Uplus and Ericsson triple fronthaul reach without degrading quality
LG Uplus says it has validated, with Ericsson, a long-distance fronthaul technology that triples the distance between radio units, the base-station antennas, and the distributed units that process the data, while keeping network quality stable. The tests covered interoperability with the operator's equipment, network automation functions and distributed-unit redundancy, with automatic failover to a backup unit when one malfunctions. According to the company, the results show that AI-RAN and cloud-based architectures can be adopted on existing facilities, which should lower deployment costs and reduce human intervention in operations. (via Seoul Economic Daily)
Smart Cities
Gangneung shows how AI integration cut emergency arrival times from 13 to 5 minutes
A UK delegation to the 2026 ITS World Congress visited the Urban Information Centre in Gangneung, South Korea, which integrates traffic monitoring, public safety, parking and automated-vehicle operations with more than 3,000 cameras and 44 staff working around the clock. The city reported a 24.2% rise in average travel speed, from 20.9 km/h to 25.9 km/h, and a drop in average junction delay from 20.9 to 17.9 seconds after optimising nine linked signal groups. Emergency-vehicle signal priority, used 7,235 times, cut average arrival times from 13 minutes to five. For the UK Department for Transport, the individual technologies are familiar; the difference lies in AI-driven integration, which moves cities' challenge from buying equipment to governing data and systems. (via Transport Technology Forum)
Xiong'an expands the "city brain" that schedules street cleaning and monitors household consumption
Xiong'an New Area, a planned city about 100 km from Beijing, runs a unified urban-management platform built on AI, big data and cloud computing that schedules street washing from weather forecasts at hundred-metre resolution and sends orders straight to crews' phones. The system also continuously monitors household water and gas use, dispatching community workers when it detects anomalies, and records recycling, passing part of the proceeds and reward points back to residents. Behind it sits the Xiong'an Urban Computing Center, with 1,500 server racks, 3,000 petaflops of AI computing power and 50 petabytes of storage. The model illustrates the efficiency potential of integrated management, but also the depth of household monitoring it presupposes. (via Bastille Post)
Seoul ranks first for AI implementation and third overall in the Observer's new city index
The Seoul Metropolitan Government announced that the capital placed third overall and first for implementation in the 2026 AI Cities Index, the first city-level ranking from Britain's The Observer, built on Tortoise Media's Global AI Index, which assessed 57 cities in 36 countries on 50 indicators of investment, innovation and implementation; San Francisco and New York lead. According to the index, while public-sector AI adoption in US hubs remains fragmented across isolated pilots, Seoul has achieved systematic integration, with a municipal committee overseeing 61 initiatives, including late-night autonomous buses, flood forecasting on 15 high-risk roads and check-ins for vulnerable residents. The city plans to train 10,000 AI professionals a year and set up a 500 billion won fund, about $377 million. (via The Korea Times)