AI Push Deepens Across Operations, Portfolios and Security as Oversight Lags

Lei Feng Net reported that Anta Group has put AI agents into production across content production, overseas ordering and cross-border returns. In Southeast Asia e-commerce, an AI video agent uses ArkClaw to collect trends and benchmark videos, Seedance 2.0 to generate product clips, and multi-site publishing tools to distribute them. New interns can produce qualified sales videos in half a day, and the system adds about 15 more videos per day at the same token consumption. For overseas orders, Anta's international division handles four templates with about 90% overlapping data. An agent automates order-number conversion, category filtering and format alignment, cutting full order decisions from half a day to 10 to 20 minutes, reducing manual entry by about 90% and raising data accuracy above 99%. It covers shoes and apparel and is planned for product-line planning and competitor price analysis. In cross-border returns, an agent pulls return data from e-commerce platforms, creates ERP documents and summarizes status. It now creates more than 1,000 return orders per month in Anta's TikTok cross-border business and is planned for FILA and other brands. Wu Yongliang, Anta's global IT delivery and support director, said the real test is whether business teams cannot work without an agent, whether workflows default to agent participation, and whether agents can be replicated across countries.

A survey of 155 North American family offices by RBC Wealth Management and Campden Wealth found that 85% rank artificial intelligence as their top investment theme, even though 75% expect an AI bubble to burst within five years. AI was the top theme for 2026 at 59%. Over the next 12 months, a combined 85% ranked AI as an asset class or investment theme likely to reward investors, ahead of U.S. equities at 76% and direct investments at 72%. Over two to five years, U.S. equities led at 69%, followed by AI at 64% and direct investments at 62%. Real estate was chosen by 26% for the next 12 months and 51% for the two-to-five-year horizon. Bill Ringham, director of private wealth strategies at RBC, said even patient capital can be caught up in fear of missing out. The survey also found 76% ranked excessive government borrowing as the top market risk likely to crystallize over five years, followed by an AI investment bubble burst at 75% and AI advancements affecting employment at 66%.

An EY report found that 50% of UK CEOs named AI as one of the three biggest contributors to productivity improvements over the past year, ahead of redesigning business processes at 40% and reskilling workers at 38%. But only 14% have full real-time visibility over AI costs, and 32% said they cannot properly assess AI's return on investment. Legacy systems, cited by 34%, and organizational complexity, cited by 29%, remain major blockers. Three in four UK CEOs expect skills shortages to worsen over the next three years, more than concerns about access to capital. Silvia Rindone, UK and Ireland managing partner at EY, said CEOs want clear evidence that investment is delivering stronger productivity and measurable business value.

A Visier report found that 48% of employees admitted exaggerating their AI usage or expertise to colleagues or managers some or most of the time, and 45% felt pressured to use AI at work. More than half, 54%, said AI had significantly changed their roles over the past two years. Thirty-two percent said their employer had no clear AI plan, while 27% were not sure their employer had any AI plan at all. Only one-third believed their leaders understood how much AI training and support they needed. Upskilling opportunities were the most sought-after change at 32%, just ahead of greater transparency at 31%. Job loss was the most frequently identified concern, raised by 24%. Dr Andrea Derler, research and customer value principal at Visier, said people want to be equipped and informed, not shielded from AI-driven change.

Obsidian Security said shadow integration has become a dangerous and often overlooked entry point for modern breaches as organizations adopt SaaS applications and AI tools. Hundreds of agents may already exist inside the average enterprise, with employees authorizing third-party OAuth applications that gain persistent access to business-critical systems. AI agents and automated workflows are being embedded into platforms such as Salesforce, ServiceNow, Workday and Slack. Obsidian said SaaS breaches have increased by 300% and projected more than one billion agents in enterprise environments by 2028. Low-code, no-code and AI platforms make it easier to deploy autonomous agents without security oversight. Obsidian disclosed a one-click remote code execution vulnerability in Flowise, an open-source platform for building AI agents and LLM workflows with more than 52,000 GitHub stars. An attacker could compromise a server by persuading an authorized user to import a malicious chatflow. Obsidian recommended visibility across users, applications, OAuth grants and permissions, and said knowledge graphs can show how people, accounts, tokens and resources relate across systems.

NPR reported that police departments across the U.S. are using real-time crime centers powered by AI software such as Fusus from Axon, which synthesizes data and lets command centers watch an entire city from one place. Flock automated license plate readers have faced backlash, but privacy experts say they are just the tip of the iceberg. Andrew Guthrie Ferguson, a law professor at George Washington University, said real-time crime centers have not just ALPR cameras and sensors but video analytics, drones and time-machine-like investigation capabilities. Axon Fusus also pulls footage and data from body cameras, dashcams, 911 dispatch, gunshot detectors and surveillance cameras, including private businesses that opt in. In 2024, 404 Media reported that just over 100 law enforcement agencies had deployed Fusus, which Axon acquired that year; Axon says the number has more than doubled since. The ACLU warned in August that Flock is far from the only vendor and that replacing it with another company's surveillance may not solve privacy and civil liberties problems. Motorola offers a similar nationwide system, the National Vehicle Location Service, through its unit Vigilant Solutions. Its commercial ALPR unit, the Digital Recognition Network, makes deals for tow trucks and other fleet vehicles to carry cameras and collect data, amassing billions of plate reads and claiming 500 million monthly with no apparent data retention policy. The data are sold commercially and made available to law enforcement for an extra charge. The article argued that if a jurisdiction allows license scanners, how the system operates matters more than which vendor provides it.

TechRadar reported that the collision between AI and the physical world is creating a new infrastructure challenge. The FAA expects the U.S. commercial drone fleet to grow from about 966,000 aircraft at the end of 2024 to around 1.18 million by 2029. An estimated 4.66 million industrial robots were working in factories worldwide in 2024, with another half a million installed during the year. Autonomous vehicles, delivery drones, agricultural machinery, warehouse robots and industrial equipment generate sensor readings, video, telemetry, diagnostics and operational logs, while also needing fresh maps, software updates, configurations, AI models and instructions. A 2019 university paper on 6G identified mobile robots, drones and autonomous vehicles as drivers of demand for ubiquitous wireless intelligence, with networks capable of supporting a 10,000-fold increase in traffic. In June, machines surpassed 57% of global web traffic for the first time, and Cloudflare predicts non-human traffic could eventually surpass human traffic by a factor of 1,000 within five years. Google DeepMind's Gemini Robotics 1.5 showed skills transferring zero-shot between physically different robots, with a task taught using a two-armed Franka robot executed by a humanoid without data collected on the target machine.

TechRadar also reported that as AI agents take over tasks, workflows and decisions, human strategy becomes more important. Companies need to define what they are trying to accomplish, what information the technology needs, how work moves between steps, and when AI can decide on its own. In marketing, an audience model could identify people likely to respond to an offer while missing customers the business wants to develop, a content agent could generate clicks but be inconsistent with brand direction, and a performance system could optimize a metric that does not reflect the larger business objective. Those are not necessarily AI failures but design failures around objectives, context and connections between decisions. The report compared the risk to two decades of enterprise technology sprawl, where CRM, marketing automation, analytics and project management systems each solved a problem but created integration work. As companies adopt AI across functions, they could recreate that problem faster with many models and agents acting on different data, instructions and outcomes. The report said requiring a person to approve every AI output would defeat the purpose of autonomous systems, so companies must decide where AI can operate independently, where monitoring is sufficient, and where human involvement is justified.

Editor's Summary The reports show AI moving rapidly into e-commerce operations, family-office portfolios, corporate productivity programs and public security systems. The common gap is governance: many organizations lack clear measurement, training, integration controls and legal limits. How quickly those controls catch up will shape whether AI's promised gains are durable or create new operational, financial and civil-liberties risks.