AI Agents Become Digital Workers as Security Flaws Drive New Industry Rules · 1025-767

2026-10-25 16:00

Author: RadarAI Editorial Editor: RadarAI Editorial Last updated: 2026-10-25 Review status: Editorial review pending Brief 速报官方 AI动态开源

Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.

## 🔍 Core Insights **AI agents** are rapidly evolving from single-point tools into 'digital employees' with autonomous decision-making capabilities, but the accompanying **security vulnerabilities** and **governance vacuum** are forcing the industry to establish new rules. Meanwhile, the competitive landscape of **open-source models** is undergoing dramatic shifts, with Chinese teams leading comprehensively in performance and ecosystem, while breakthroughs in **embodied intelligence** and **full-duplex voice** signal a fundamental shift in the human-computer interaction paradigm [22][7][10]. ## 🚀 Key Developments - **OpenAI Codex fully rolled out, usage quota reset** [0][5]: Codex is now available to all users, and with the quota reset, developers can freely conduct high-intensity coding experiments. - **University of Toronto and Google paper: AI assistant choices may become the 'last major decision'** [1]: The research proposes that future decisions will be delegated to AI, and calls for establishing a unified GEO benchmark to assess the risk of AI answers being manipulated. - **Nanjing University and others open-source OpenMobile-2 mobile agent** [2]: Built on the Qwen architecture, the 27B version scores 79.9 on AndroidWorld, with the complete training environment and data made public. - **Google DeepMind research: Proactive AI in teams triggers dual perceptions of 'savior and surveillance'** [4]: Agents ran in over 20 teams for 5 months, and their unsolicited actions (such as mislabeling bugs or sharing drafts prematurely) caused stress for some employees. - **Block releases Dessa full-duplex voice model** [7][21]: Trained purely on voice with no text backbone, it supports real-time local operation on MacBook Pro M4 Max and can simulate emotions such as laughter and choking up. - **Microsoft open-sources Agensh decentralized multi-agent coding framework** [13]: Removes the central orchestrator, supporting scaling to 1,024 agents that claim tasks and merge code autonomously. - **AI vulnerability discovery accelerating, government website outages may recur** [6]: Security experts warn that AI can develop attack programs within hours, breaking the defender's patching time gap, and PHP vulnerability-type incidents are almost certain to happen again. - **Major US banks develop emergency plans for AI agent financial operations** [16]: Morgan Stanley, Bank of America, and others are addressing disaster scenarios caused by agents operating autonomously or even calling customer service, warning of low-cost deposit outflow risks. ## 🔗 Sources [0] OpenAI usage quota reset to 100% with clock adjustment — https://aihot.news/items/y08641zgkvhq3m3wam7ye5e9t [1] University of Toronto and Google paper: AI assistant choices may become the last major decision — https://aihot.news/items/nvfuzr06xq5p3otd9koo5tpjp [2] Nanjing University and other teams open-source OpenMobile-2 mobile agent, 27B version scores 79.9 on AndroidWorld — https://aihot.news/items/l52ssugscix8mf4id6wr7hmlk [4] Google DeepMind paper: Proactive AI agents enter teams, some see them as saviors, others feel surveilled — https://aihot.news/items/kvbeuuzi3ly6rdgtpcsqtj7f0 [5] OpenAI's Codex has been fully rolled out — https://aihot.news/items/emnw0127sawwqu8f2qd6ru6s0 [6] Cybersecurity experts: AI discovering vulnerabilities faster, government website outage-type incidents may recur — https://aihot.news/items/uw3wr0pwd9hw7d6i3d3t6

AI agents are rapidly evolving from single-point tools into 'digital employees' with autonomous decision-making capabilities, but the accompanying security vulnerabilities and governance vacuum are forcing the industry to establish new rules. Meanwhile, the competitive landscape of open-source models is undergoing dramatic shifts, with Chinese teams leading comprehensively in performance and ecosystem, while breakthroughs in embodied intelligence and full-duplex voice signal a fundamental shift in the human-computer interaction paradigm [22][7][10].

🚀 Key Developments

OpenAI Codex fully rolled out, usage quota reset [0][5]: Codex is now available to all users, and with the quota reset, developers can freely conduct high-intensity coding experiments.
University of Toronto and Google paper: AI assistant choices may become the 'last major decision' [1]: The research proposes that future decisions will be delegated to AI, and calls for establishing a unified GEO benchmark to assess the risk of AI answers being manipulated.
Nanjing University and others open-source OpenMobile-2 mobile agent [2]: Built on the Qwen architecture, the 27B version scores 79.9 on AndroidWorld, with the complete training environment and data made public.
Google DeepMind research: Proactive AI in teams triggers dual perceptions of 'savior and surveillance' [4]: Agents ran in over 20 teams for 5 months, and their unsolicited actions (such as mislabeling bugs or sharing drafts prematurely) caused stress for some employees.
Block releases Dessa full-duplex voice model [7][21]: Trained purely on voice with no text backbone, it supports real-time local operation on MacBook Pro M4 Max and can simulate emotions such as laughter and choking up.
Microsoft open-sources Agensh decentralized multi-agent coding framework [13]: Removes the central orchestrator, supporting scaling to 1,024 agents that claim tasks and merge code autonomously.
AI vulnerability discovery accelerating, government website outages may recur [6]: Security experts warn that AI can develop attack programs within hours, breaking the defender's patching time gap, and PHP vulnerability-type incidents are almost certain to happen again.
Major US banks develop emergency plans for AI agent financial operations [16]: Morgan Stanley, Bank of America, and others are addressing disaster scenarios caused by agents operating autonomously or even calling customer service, warning of low-cost deposit outflow risks.

🔗 Sources

[0] OpenAI usage quota reset to 100% with clock adjustment — https://aihot.news/items/y08641zgkvhq3m3wam7ye5e9t [1] University of Toronto and Google paper: AI assistant choices may become the last major decision — https://aihot.news/items/nvfuzr06xq5p3otd9koo5tpjp [2] Nanjing University and other teams open-source OpenMobile-2 mobile agent, 27B version scores 79.9 on AndroidWorld — https://aihot.news/items/l52ssugscix8mf4id6wr7hmlk [4] Google DeepMind paper: Proactive AI agents enter teams, some see them as saviors, others feel surveilled — https://aihot.news/items/kvbeuuzi3ly6rdgtpcsqtj7f0 [5] OpenAI's Codex has been fully rolled out — https://aihot.news/items/emnw0127sawwqu8f2qd6ru6s0 [6] Cybersecurity experts: AI discovering vulnerabilities faster, government website outage-type incidents may recur — https://aihot.news/items/uw3wr0pwd9hw7d6i3d3t6

← Back to Updates