PulseAugur
中
实时 17:07:20
English(EN) Open and Emergent Problems in Agentic Privacy and Security: A Contextual Angle

新研究和研讨会详细介绍了AI代理的安全和隐私风险

多篇研究论文和一份Google研讨会报告强调了增强自主AI代理的安全和隐私措施的迫切需求。这些工作解决了诸如非结构化输入歧义、概率性控制流以及代理可能越界或未能执行基本安全任务等挑战。提出的解决方案包括诸如用于证据溯源的Proof-of-Use (PoU)框架、主动保证周期以及分层权限模型,以确保代理在定义的道德和安全参数内运行。 AI

影响 强调了当前AI代理中关键的安全和隐私差距,推动了对新保证框架和治理模型的研究。

排序理由 多篇arXiv论文和一份Google Research研讨会报告详细介绍了AI代理新兴的安全和隐私挑战。

在 Google AI / Research 阅读 →

AI 生成摘要 · Google Gemini · 来自 133 个来源。 我们如何撰写摘要 →

新研究和研讨会详细介绍了AI代理的安全和隐私风险

本文如何被排名

Signal score
0 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Research
多篇arXiv论文和一份Google Research研讨会报告详细介绍了AI代理新兴的安全和隐私挑战。
Source corroboration
133 independent sources
Strong cross-source corroboration — multiple independent publishers covered this within the clustering window.
Topics
safety, policy, paper
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
47 days old
Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.
Coverage growth since scoring
+66 source(s) since last score
New sources have picked up this story since our last re-score. Score will update on the next scoring pass.

完整方法见我们的编辑标准。

报道来源 [133]

  1. Google AI / Research TIER_1 English(EN) ·

    Agentic 隐私与安全中的开放性与涌现性问题:一个情境化视角

    Education Innovation

  2. arXiv cs.MA (Multiagent) TIER_1 English(EN) · Jason Patrick ·

    可验证权威性:一种用于人类、组织和代理交易的可验证权威解析与控制协议

    Payment systems authenticate users, devices, accounts, and cryptographic credentials, but a relying party, such as a counterparty, payment rail, marketplace, or autonomous agent runtime, generally cannot obtain compact, independently verifiable evidence that the party presenting …

  3. arXiv cs.LG TIER_1 English(EN) · Jesse C. Cresswell ·

    可信赖的人工智能必须考虑交互

    arXiv:2504.07170v3 Announce Type: replace Abstract: Trustworthy AI encompasses many aspirational aspects for aligning AI systems with human values, including fairness, privacy, robustness, explainability, and uncertainty quantification. The ultimate goal of Trustworthy AI researc…

  4. arXiv cs.AI TIER_1 English(EN) · Ahammod Munim Ramil, Khalid Al Kafi, M. A. Abdullahis Sany, Muktadirul Islam Nibir, Md Yeasin Ali, Mohammad Zavid Parvez, Md Sadek Ferdous ·

    人工智能中的“我”是谁?为自主代理建立去中心化身份和问责制

    arXiv:2610.17126v1 Announce Type: cross Abstract: The rapid proliferation of autonomous AI agents in decentralized ecosystems challenges existing identity paradigms. While human identity is traditionally anchored in biological and/or self-asserted attributes and government creden…

  5. arXiv cs.AI TIER_1 English(EN) · Fasheng Miao, Zhaoyu Hu, Xiang Li, Changqing An, Jiutian Zeng, Longtao Huang, Haixin Duan, Jilong Wang ·

    ProtoArena:AI代理能否发现并利用真实世界的跨组件协议漏洞?

    arXiv:2610.17837v1 Announce Type: cross Abstract: Large language model (LLM) agents are increasingly used for vulnerability analysis, yet existing benchmarks provide limited coverage of exploitation that depends on interactions among network protocol components. Such vulnerabilit…

  6. Hugging Face Daily Papers TIER_1 English(EN) ·

    ProtoArena:AI代理能否发现并利用真实世界的跨组件协议漏洞?

    Large language model (LLM) agents are increasingly used for vulnerability analysis, yet existing benchmarks provide limited coverage of exploitation that depends on interactions among network protocol components. Such vulnerabilities arise from inconsistencies in message interpre…

  7. arXiv cs.AI TIER_1 English(EN) · Subhabrata Majumdar, Rajlakshmi Chavan ·

    Stackelberg AI安全模型中的防御充分性

    arXiv:2610.09892v2 Announce Type: replace-cross Abstract: Feedback from automated testing, human red teaming, and incident response can strengthen an AI system's defenses when discovered failures lead to effective repairs. We study when this feedback process provides sufficient p…

  8. arXiv cs.AI TIER_1 English(EN) · Abbas Raftari ·

    从被动防御到主动保障:来自 OpenAI、Anthropic 和 Google Agent Security 事件的经验教训

    arXiv:2610.12463v1 Announce Type: cross Abstract: In 2026, cybersecurity evaluations involving OpenAI, Anthropic, and Google agents reached real systems outside their authorized test scope. The paths were different. OpenAI agents exploited research infrastructure, coordinated acr…

  9. arXiv cs.AI TIER_1 English(EN) · Gaston Besanson ·

    概率感知,确定性权威:将模型生成的观测值纳入充分性检查的治理合同

    arXiv:2610.10978v1 Announce Type: cross Abstract: When a field that an authority contract needs exists only in unstructured evidence, a model can sense it. We admit the model's output only as an observation record with a score. An admission policy, with thresholds fitted on a hel…

  10. arXiv cs.AI TIER_1 English(EN) · Youwei Feng, Yitong Zhang, Yuetong Liu, Jia Li ·

    仅靠安全操作无法确保安全代理:使用 Guard 模型识别未履行的义务

    arXiv:2610.11773v1 Announce Type: new Abstract: Guard models are increasingly used to safeguard LLM-based agents, primarily by identifying actions that agents are forbidden to perform. However, identifying forbidden actions alone is insufficient to ensure agent safety. In this pa…

  11. arXiv cs.AI TIER_1 English(EN) · Shengjie Ma, Chenlong Deng, Jiaxin Mao, Jiadeng Huang, Teng Wang, Junjie Wu, Changwang Zhang, Jun wang ·

    使用证明:缓解深度研究代理中的工具调用黑客攻击

    arXiv:2510.10931v3 Announce Type: replace Abstract: While reinforcement learning (RL) enhances their ability to plan and reason across retrieval steps, we identify a critical failure mode in this setting: Tool-Call Hacking. Unlike execution-based tools (e.g., code or math), whose…

  12. arXiv cs.LG TIER_1 English(EN) · Leo Zeitler, Jack Richings, Victoria Nockles ·

    AI Safety Considerations for Agents With Limited Time to Act

    arXiv:2610.10285v1 Announce Type: cross Abstract: In the wake of the increasingly public discussion about AI alignment, recent work has tried to propose specific AI architectures that behave safely. However, the proposed arguments that seemingly demonstrate proved alignment mostl…

  13. arXiv cs.AI TIER_1 English(EN) · Venkata M Sangaraju, Sudhir Vissa ·

    面向企业AI代理的隐私保护列级访问控制:一种基于 lineage 的派生门控框架

    arXiv:2610.07258v1 Announce Type: cross Abstract: Enterprise AI agents that share a memory store face two unaddressed risks: sensitive data can leak through legitimately computed results the requester could not derive, and departments can silently compute a same-named key perform…

  14. arXiv cs.AI TIER_1 English(EN) · Jinhao Zhu, Xiao Huang, Kevin Tseng, Gil Vernik, Shishir G. Patil, Vivian Fang, Raluca Ada Popa ·

    MiniScope:以最小权限授权代理

    arXiv:2512.11147v2 Announce Type: replace-cross Abstract: AI agents are increasingly granted autonomous access to sensitive user data and third-party services, making effective permission management a critical security challenge. Existing permission models, however, typically rel…

  15. arXiv cs.AI TIER_1 English(EN) · Shaswata Mitra, Raj Patel, Subash Neupane, Sudip Mittal, Md Rayhanur Rahman, Shahram Rahimi ·

    规则终结与法官始于何处:衡量多智能体系统安全中的判断边界

    arXiv:2610.07657v1 Announce Type: new Abstract: LLM-based multi-agent systems (MAS) engage tools, share memory, and delegate tasks, often encountering adversarial content. Current defenses for MAS are typically evaluated in isolation, focusing on one attack type at a time, which …

  16. arXiv cs.MA (Multiagent) TIER_1 English(EN) · Shahram Rahimi ·

    规则终结与法官始处:衡量多智能体系统安全中的判断边界

    LLM-based multi-agent systems (MAS) engage tools, share memory, and delegate tasks, often encountering adversarial content. Current defenses for MAS are typically evaluated in isolation, focusing on one attack type at a time, which can lead to costly and hard-to-audit outcomes. T…

  17. Hugging Face Daily Papers TIER_1 English(EN) ·

    人工智能是帮助网络攻击者还是防御者?来自非公开漏洞和后续攻击的证据

    The release decision for frontier AI systems increasingly relies on cyber capability benchmarks, yet public vulnerability benchmarks can expose agents to previously published advisories, exploits, and fixes, making it difficult to distinguish prior exposure from capability on uns…

  18. arXiv cs.MA (Multiagent) TIER_1 English(EN) · Anantika Mannby ·

    拓扑条件后门:语言模型在推断其处于多智能体系统时插入漏洞

    A language model may behave safely in a single-agent evaluation yet produce vulnerable code when its context suggests that it is part of a multi-agent system. We study this failure mode by fine-tuning Qwen2.5-7B-Instruct to condition code generation on deployment topology inferre…

  19. arXiv cs.AI TIER_1 English(EN) · Neeraj Karamchandani, Piyush Nagasubramaniam, Xinhong Xie, Sencun Zhu, Dinghao Wu ·

    Agent-Security基准中的威胁保持表示敏感性

    arXiv:2610.03585v1 Announce Type: cross Abstract: Security benchmarks for LLM-based agents often report the attack success rate (ASR) as a measure of model robustness and use these scores to compare different models and defense mechanisms, assuming that they describe the security…

  20. Hugging Face Daily Papers TIER_1 English(EN) ·

    面向企业AI代理的隐私保护列级访问控制:一种基于 lineage 的派生门控框架

    Enterprise AI agents that share a memory store face two unaddressed risks: sensitive data can leak through legitimately computed results the requester could not derive, and departments can silently compute a same-named key performance indicator (KPI) through conflicting logic. Ex…

  21. AWS Machine Learning Blog TIER_1 English(EN) · Rui Li ·

    构建AI代理的信任:使用Amazon Bedrock AgentCore的生产就绪评估框架

    Autonomous AI agents fail softly: latency stays flat while answer quality quietly erodes. This post shares a nine-stage evaluation framework, built on Amazon Bedrock AgentCore, that catches silent quality regressions across the full agent lifecycle, from first commit to live prod…

  22. Forbes — Innovation TIER_1 English(EN) · Tim Keary, Contributor ·

    为什么运行时安全可能成为防御 AI Agent 的关键

    As AI agents threaten to accelerate vulnerability discovery and exploitation, Sweet Security turns to virtual patching to help mitigate exploits at machine speed.

  23. Forbes — Innovation TIER_1 English(EN) · Gopichand Mannava, Forbes Councils Member ·

    科技领袖们设计AI代理以防止错误的七种方式

    The biggest breakthrough isn’t making AI agents more autonomous. It’s teaching them to doubt themselves.

  24. Forbes — Innovation TIER_1 English(EN) · Ewan Spence, Senior Contributor ·

    Pixel 11 Pro:为何 Agentic AI 无法掩盖其平庸的硬件升级

    Google’s Pixel 11 Pro pairs an ambitious agentic AI vision with iterative hardware updates. Is this enough for a workhorse phone for everyday use?

  25. Forbes — Innovation TIER_1 English(EN) · Steve Karp, Forbes Councils Member ·

    构建还是购买AI代理?优劣势与复杂性在此

    Whether you build or buy AI agents, be sure the choice is driven by value above all else.

  26. Forbes — Innovation TIER_1 English(EN) · Etay Maor, Forbes Councils Member ·

    AI 代理无法区分指令和数据——身份验证也无法解决此问题

    An attacker can spoof a location or clone a recognized-looking device. What's harder to fake is behavior.

  27. Forbes — Innovation TIER_1 English(EN) · Srinivas Shekar, Forbes Councils Member ·

    敲响警钟,集结队伍:AI 攻击你公司的风险真实存在

    With AI hacking becoming more severe and urgent, there are steps you can take to mitigate your exposure.

  28. Forbes — Innovation TIER_1 English(EN) · Tim Keary, Contributor ·

    为何攻击路径分析可以阻止恶意AI代理蜂群

    Attack path analysis could potentially provide a solution to stop the next rogue AI agent breach according to one startup.

  29. Forbes — Innovation TIER_1 English(EN) · Ashoka, Contributor ·

    人工智能成为新的审查工具:监控与开放模型的论证

    A conversation with social entrepreneur Sascha Meinrath about privacy breaches and censorship disguised as AI safeguards, and what can be done about it.

  30. Forbes — Innovation TIER_1 English(EN) · Vikram Pande, Forbes Councils Member ·

    人工智能时代的零信任:为何身份必须成为企业控制平面

    Most enterprises did not design their identity architectures for a world in which software could act on behalf of people, yet that world is arriving quickly.

  31. Forbes — Innovation TIER_1 English(EN) · Tony Grout, Forbes Councils Member ·

    可信AI始于可信数据:为何治理必须为Agentic时代演进

    Trusted AI doesn’t begin with the model. It begins with trusted data.

  32. dev.to — Claude Code tag TIER_1 English(EN) · Dwayne McDaniel ·

    Claude 代码、Cursor 和 Codex 的 AI Hook:在代理使用它们之前阻止秘密

    <p><a class="article-body-image-wrapper" href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2F2fwbyrwyd5wvol385p6d.png"><img alt="AI Coding Assist…

  33. dev.to — Claude Code tag TIER_1 English(EN) · Maria ·

    AI 编码代理会泄露你的 API 密钥吗?

    <p>In January <a href="https://www.theregister.com/2026/01/28/claude_code_ai_secrets_files/" rel="noopener noreferrer">The Register</a> showed Claude Code 2.1.12 reading a <code>.env</code> it had been told to ignore and printing the secrets in it. While I was working on my new M…

  34. The Verge — AI TIER_1 English(EN) · Hayden Field ·

    人工智能代理制造商承诺保护隐私——他们会兑现承诺吗?

    At this year's OpenAI DevDay, CEO Sam Altman unveiled the company's new AI agent Dots - and told the crowd that the company wants to "set a new standard for privacy in frontier AI." OpenAI would spend the day taking veiled shots at Meta's Muse, its primary competitor, for failing…

  35. Fortune TIER_1 English(EN) · Nick Lichtenberg ·

    CLEAR创始人:当我们修复身份层时,快速AI也可以是安全的AI

    Every AI agent borrows its authority from a person. Before agents act for us, let’s guarantee you are you.

  36. AI Business TIER_1 English(EN) · Scarlett Evans ·

    IBM:Agentic AI 正在重塑治理、控制和信任

    As AI agents move from answering questions to making decisions and acting, businesses face a growing challenge of keeping accountability in step with the pace of change.

  37. AI Business TIER_1 English(EN) · Esther Shittu ·

    DeepSeek 以及企业应从 AI 代理漏洞中学到什么

    A new case of an out-of-control AI model shows that enterprises should treat models and agents as toddlers that need supervision and explicit instructions.

  38. dev.to — MCP tag TIER_1 English(EN) · ScriptMasterLabs ·

    AI 代理能否安全访问扩展访问的患者数据?已识别 5 项能力,0 项写入

    <h1> Can AI Agents Safely Access Expanded Access Patient Data? 5 Capabilities Counted, 0 Write </h1> <p>Yes — on MedaSystems' new MCP server the safe answer is structural, not scored: the server is read-only, it inherits the signed-in user's own role-based permissions, and it is …

  39. dev.to — MCP tag TIER_1 English(EN) · Ahmad Karawash ·

    Amoxi:能够重启后存活并与其他AI代理对话的AI代理

    <p><em>Durable runs that pick up where they stopped, side effects that never happen twice, and your agents offered to Claude, ChatGPT and other agents over MCP and A2A.</em></p> <p>In article 5 every run took about a second. Real agent work is slower: several model calls, a payme…

  40. dev.to — MCP tag TIER_1 English(EN) · ScriptMasterLabs ·

    AI 代理能否安全地订购网络连接?已识别 8 种能力,其中 5 种涉及采购

    <h1> Can AI Agents Safely Order Network Connections? 8 Capabilities Counted, 5 of Them Procurement </h1> <p>Yes — if ordering stays in the human-confirm band and only discovery clears automatically, because on DE-CIX's new MCP server the write surface is not editing or publishing…

  41. Medium — AI coding tag TIER_1 English(EN) · Asad Khan ·

    Agentic Test Automation: Why the Agent Must Never Grade Its Own Test

    <div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/agents-under-test/agentic-test-automation-why-the-agent-must-never-grade-its-own-test-774856a6ea93?source=rss------ai_coding-5"><img src="https://cdn-images-1.medium.com/max/1376/1*HuSu1R1iKHDE…

  42. dev.to — MCP tag TIER_1 English(EN) · Guangyi Zhao ·

    让 AI 代理驾驶 Ghidra:REA 170 MCP 工具、一次恢复的密码以及局限性的实际测试

    <p>Author: Guangyi Zhao<br /> Date: 2026-10-13<br /> Subject: morluto/rea (REA: Reverse Engineer Anything), npm package rea-agents 6.4.0, MIT license<br /> Environment: Ubuntu 24.04 (Linux x64), Node 24.20; target is an ELF built from my own C source</p> <h2> 1. Problem and backg…

  43. dev.to — MCP tag TIER_1 English(EN) · ScriptMasterLabs ·

    AI代理能否安全管理物理设备?已识别10个能力组,4个仅限更改

    <h1> Can AI Agents Safely Manage Physical Devices? 10 Capability Groups Counted, 4 Change-Only </h1> <p>Yes — if the tools that can change a device are withheld by a switch the agent itself cannot flip, and destructive commands stay in the human-confirm band even after that switc…

  44. dev.to — MCP tag TIER_1 English(EN) · Bhanu Nagpure ·

    漏洞在于编排:查找并切断AI代理的数据泄露路径

    <p>Give an agent three ordinary tools: one that reads a GitHub issue, one that<br /> reads a local folder, one that can post a GitHub comment. Each behaves exactly<br /> as documented. Each is useful on its own. Together they form an exfiltration<br /> path — an instruction hidde…

  45. dev.to — MCP tag TIER_1 Français(FR) · iamthat ·

    drift: 持续的AI代理强制执行指令

    <p>🛡️ drift — continuous mandate enforcement for AI agents</p> <p>An agent is granted a mandate: who to pay, how much, in which currency, until when.<br /> AP2 verifies that mandate once, at signature time. After that the agent acts alone.</p> <p>This is not theoretical. In Googl…

  46. dev.to — MCP tag TIER_1 English(EN) · ScriptMasterLabs ·

    AI 代理能否安全访问您的商务电话系统?Viirtue MCP 经过计数和网关测试

    <h1> Can AI agents access your business phone system safely? </h1> <p>Yes — if the MCP server is read-first, fails closed on recorded content, and exposes no destructive tools. Viirtue MCP, launched October 15, 2026, connects Claude, ChatGPT and Codex to a business phone system a…

  47. Medium — AI coding tag TIER_1 English(EN) · Old Developer ·

    你的 AI 代理无法隐藏软件工程的复杂性

    <div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@iyakin.yz/your-ai-agent-cant-hide-the-complexity-of-software-engineering-4b67b8823eaf?source=rss------ai_coding-5"><img src="https://cdn-images-1.medium.com/max/1774/1*0dZSINd_4vLXH5DZViTCTg.p…

  48. dev.to — MCP tag TIER_1 English(EN) · Jorge Peraza ·

    AI Agent 安全保障:计划审批与 MCP 规范化

    <p><strong>TL;DR</strong></p> <ul> <li> <strong>The vulnerability:</strong> An October 7, 2026 security report identified eight vulnerabilities across seven AI coding agents, exposing attack vectors like git configuration hijacking, unsafe plugin execution, and sandbox escapes ca…

  49. Medium — Claude tag TIER_1 English(EN) · David Wainaina ·

    DevOps工程师入门AI代理

    <div class="medium-feed-item"><p class="medium-feed-snippet">AI agents are natural next steps for DevOps Engineers because they can do what you already know how to do, in a fraction of the time. This&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@wainaina/get…

  50. dev.to — MCP tag TIER_1 English(EN) · RevOS ·

    AI 代理身份:谁在行动,以及在谁的授权下?

    <p>Suppose you tell an agent: <em>move Acme to At Risk, create a task for the account owner, and prepare a renewal plan.</em> The agent does exactly that.</p> <p>What should your CRM say happened?</p> <p><strong>Last modified by Alex</strong>, because you asked for it? <strong>La…

  51. Mastodon — sigmoid.social TIER_1 English(EN) · [email protected] ·

    🤖 深度信任:在不失控的情况下赋予代理更多自主权 企业领导者希望AI代理承担更多工作,但又不完全信任它们会协同工作

    🤖 Trust in Depth: Giving agents more autonomy without losing control Enterprise leaders want AI agents to take on more work but don't fully trust them to act correctly every time. Trust in Depth is AWS's layered approach-scope, know, bound, observe, and prove-that l... 📰 Source: …

  52. Medium — Claude tag TIER_1 English(EN) · Cyber Chronicle ·

    Claude Managed Agents:AI安全领域的新攻击面

    <div class="medium-feed-item"><p class="medium-feed-image"><a href="https://meetcyber.net/claude-managed-agents-the-new-attack-surface-in-ai-security-49b8166fdfe1?source=rss------claude-5"><img src="https://cdn-images-1.medium.com/max/1536/1*hMyrxbL0S2vwF2Tiqs5DPA.png" width="153…

  53. Towards AI TIER_1 English(EN) · Arun Prabhakar ·

    代理身份保证——企业采信可靠AI的基础要素

    <figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bcr2m6ySssYWJlo6PxLiYw.png" /><figcaption>Image generated by AI</figcaption></figure><p>As AI Agents increasingly act on behalf of users and within enterprise applications and systems, a distinct and governable i…

  54. dev.to — MCP tag TIER_1 English(EN) · levl ·

    让你的AI代理跟随Polymarket交易员:Claude Code, Codex, Cursor和MCP

    <p>When a Polymarket trader I follow makes a trade, I want an AI agent to see it, look at the market, and decide whether to do anything about it.</p> <p>Agents like Claude Code or Codex can read a fill and reason about a market, but they don't work well as long-running listeners.…

  55. Towards AI TIER_1 English(EN) · PhynixAI ·

    可靠AI代理背后的安全架构

    <div class="medium-feed-item"><p class="medium-feed-image"><a href="https://pub.towardsai.net/the-security-architecture-behind-reliable-ai-agents-e8802ca16717?source=rss----98111c9905da---4"><img src="https://cdn-images-1.medium.com/max/1024/1*bu9k9dGNtZwv7JkySO_TFw.jpeg" width="…

  56. Medium — AI coding tag TIER_1 English(EN) · Cyberathon Technology ·

    超越AI编码:如何在现代软件开发中保障AI自动化

    <div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@contact.cyberathon/beyond-ai-coding-how-to-secure-ai-automation-in-modern-software-development-80f64043dd26?source=rss------ai_coding-5"><img src="https://cdn-images-1.medium.com/max/2600/1*yd…

  57. dev.to — MCP tag TIER_1 English(EN) · Auth By Example ·

    当AI代理代表他人行事时应记录什么

    <p>Most agent stacks already log plenty: prompts, completions, tool calls, latency, token counts. That's useful for debugging. After an incident, though, a security reviewer will ask something the traces can't answer: was this action allowed, for this agent, on behalf of this per…

  58. dev.to — MCP tag TIER_1 English(EN) · Pangolinfo ·

    如何将AI代理连接到实时Amazon数据:完整可运行指南

    <p>An AI agent is, at minimum, a language model wrapped in a loop: it observes, reasons, acts, and observes again. What distinguishes a data-connected agent from a bare model is a single property — grounding. A grounded agent can resolve a claim against an external source at the …

  59. Mastodon — sigmoid.social TIER_1 English(EN) · [email protected] ·

    AI代理用于验证链上声明的清单:固定调用,将空的0x视为不可验证,拒绝未知单位,将收据绑定到获取的对象,以及

    A checklist an AI agent uses to verify onchain claims: pin the call, treat empty 0x as unverifiable, refuse unknown units, bind receipts to fetched objects, and treat finality as part of the number. # web3 # crypto # ai # programming # software # coding # development # engineerin…

  60. Mastodon — sigmoid.social TIER_1 English(EN) · [email protected] ·

    AI Agents with Python 2026 — 第9部分上线!我们如何使AI代理更安全、更可靠?探索:提示注入 工具安全 护栏 验证 Hum

    AI Agents with Python 2026 — Part 9 is LIVE! How do we make AI agents safer and more reliable? Explore: Prompt injection Tool security Guardrails Validation Human approval Failure handling Observability The model proposes. Code enforces. Get the codes: https:// github.com/ABKarik…

  61. Towards AI TIER_1 English(EN) · Devi ·

    AI Agent背后的隐藏架构问题

    <h4>What Satya Nadella’s recent essay and research on agent failures reveal about building trustworthy AI architecture</h4><p>Satya Nadella’s <a href="https://x.com/satyanadella/status/2108931348857827686?s=20">recent article </a>on X, makes a deliberately uncomfortable argument:…

  62. dev.to — MCP tag TIER_1 English(EN) · InApp ·

    brick.blue:一个AI代理用代币换取金钱的交易所

    <p>I run a paid API gateway (x402.freeq.one) — about 32 tools agents can call, settled in USDC on Base. The supply side was never my problem. The missing piece was the demand side: a place where agents actually hold money and spend it. Last week I found <a href="https://brick.blu…

  63. dev.to — MCP tag TIER_1 Română(RO) · Jonathan Blessing ·

    AI Operator Security

    <p>In September, at MCPCon Europe in Amsterdam,<sup id="fnref1">1</sup> most of the security talks were about the same thing: protecting you from someone else's AI. Prompt attacks. Poisoned models. Rogue agents. The usual framing: someone else's AI brings the harm.</p> <p>But the…

  64. Medium — MCP tag TIER_1 English(EN) · Rudolph Koegelenberg ·

    构建AI代理正变得越来越容易。让它们值得信赖才是真正的工作。

    <div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/rudolphk/building-ai-agents-is-getting-easier-making-them-trustworthy-is-the-real-work-4c386065b19f?source=rss------mcp-5"><img src="https://cdn-images-1.medium.com/max/1672/1*QRJmUFLywkddRCU5k…

  65. dev.to — MCP tag TIER_1 English(EN) · Roy Berris ·

    为 AI 代理设计 API:模式、安全和 MCP

    <p><strong>TL;DR:</strong> <a href="https://www.postman.com/state-of-api/2025/" rel="noopener noreferrer">Postman's 2025 State of the API report</a> shows that 89% of developers use AI in their daily work, but only 24% design APIs with AI agents in mind. My answer is to design fo…

  66. Mastodon — sigmoid.social TIER_1 Italiano(IT) · [email protected] ·

    Agentic AI:ICO 信号其数据保护优先事项 #AgenticAI #AgenticArtificialIntelligence #AI #ArtificialIntelligence

    https://www. europesays.com/3302755/ Agentic AI: ICO signals its data protection priorities # AgenticAI # AgenticArtificialIntelligence # AI # ArtificialIntelligence

  67. Towards AI TIER_1 English(EN) · JingJing (Chris) Bao ·

    OAuth 2.0 Token Exchange — 为 AI 代理授权奠定基础

    <h3>Background</h3><p>In a previous article, I walked through the three most common OAuth 2.0 grant flows: Authorization Code Flow, PKCE Flow, and Device Flow. We looked at which scenarios and application types each one fits, what problems they solve, and how they differ in imple…

  68. dev.to — MCP tag TIER_1 English(EN) · developerz.ai ·

    使用 db-mcp-gateway 为 AI 代理提供安全数据库访问

    <h1> Secure Database Access for AI Agents with db-mcp-gateway </h1> <h2> Introduction </h2> <p>Artificial intelligence agents increasingly need to read data from production databases to generate accurate responses. Providing that access without exposing credentials is a hard prob…

  69. dev.to — MCP tag TIER_1 English(EN) · ScriptMasterLabs ·

    Apollo GraphOS Agent 服务:Apollo 同意代理需要明确规则 — 但治理仅限于访问权限

    <h1> Apollo GraphOS Agent Services: Apollo Agrees Agents Need Explicit Rules — But Governance Stops at Access </h1> <p>On October 7, 2026, Apollo GraphQL launched <strong>GraphOS Agent Services</strong> at Apollo Summit in San Francisco: search, identity, policy, and audit servic…

  70. Medium — Claude tag TIER_1 English(EN) · Cywarden ·

    2026年人工智能网络安全指南

    <div class="medium-feed-item"><p class="medium-feed-image"><a href="https://cywarden.medium.com/the-2026-guide-to-ai-cybersecurity-73147d5aa3b6?source=rss------claude-5"><img src="https://cdn-images-1.medium.com/max/1672/1*yCu9vCLeOuUSuSy2k6x14A.png" width="1672" /></a></p><p cla…

  71. dev.to — MCP tag TIER_1 English(EN) · Umang Kumar ·

    用于 AI 代理操作的哈希链式审计日志,仅 40 行

    <p>When an AI agent calls a tool — reads a file, applies a manifest, installs a package — you want an audit log you can trust weeks or months later. Tamper-evident doesn't require a blockchain or a signing service. A SHA-256 hash chain over append-only entries is enough to detect…

  72. dev.to — MCP tag TIER_1 English(EN) · Umang Kumar ·

    为什么批准提示词无法作为编码代理的安全边界

    <p>When a coding agent asks a human to approve a file change, a database call, or a deploy, the approval prompt feels like a security boundary. It's not.</p> <p>In practice, approval prompts leak authority for three reasons: approval fatigue, approvals that never expire, and stal…

  73. dev.to — MCP tag TIER_1 English(EN) · Umang Kumar ·

    阻止读取再泄露链:AI代理的会话污点

    <p>Most AI agent incidents people worry about are not one bad tool call. They are two ordinary ones in a row.</p> <ol> <li>The agent reads a file. Maybe it is <code>.env</code>, maybe <code>config/credentials.yml</code>, maybe a token pasted into a ticket it was asked to summaris…

  74. dev.to — MCP tag TIER_1 English(EN) · Mike Kim ·

    让团队的AI在共享计算机上运行命令:两层权限和审计日志

    <p>Once ChatGPT or Claude can edit files and run commands on your computer through MCP, using it alone is easy. Sharing it with a team raises questions fast:</p> <ul> <li>Who asked the AI to do what, on which computer?</li> <li>What if a command the AI ran had an API key in it?</…

  75. Towards AI TIER_1 English(EN) · Sneha Padgaonkar ·

    如果恩佐·法拉利因其汽车所为而受到指责?AI代理授权问题

    <h4>AI agents can now make decisions, call tools, change systems, and act on their own. When something goes wrong, who is actually responsible?</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*X7_ssPbLBeHDLV-1itIWZw.png" /><figcaption>The Ferrari problem: w…

  76. dev.to — MCP tag TIER_1 English(EN) · mech.app ·

    MCP 代理间漏洞:谷歌的缺陷揭示了协议级安全边界的哪些问题

    <p>The Model Context Protocol (MCP) is becoming the default wire format for agent-to-agent communication. Anthropic designed it, major vendors adopted it, and now security researchers have found structural flaws in how it handles trust boundaries when one agent calls another agen…

  77. dev.to — MCP tag TIER_1 English(EN) · Azrty ·

    AI代理安全风险:2026年的4项控制措施

    <p>Agents that issue refunds and reroute shipments are privileged users that read hostile text. Here are the 2026 incidents, the four controls that make them safe, and the questions to put to any vendor.</p> <p>AI agent security risks come down to one structural problem: an agent…

  78. Towards AI TIER_1 English(EN) · Sumit Verma ·

    删除是一种承诺:安全研究员眼中的 AI 对话隐私

    <h4>An OpenAI privacy bug in the gap between an object and the assets derived from it</h4><h3>Summary</h3><p>When you share a ChatGPT conversation, the share page gets an Open Graph preview image served from a separate hostname. I found that deleting the conversation correctly in…

  79. Medium — MLOps tag TIER_1 English(EN) · Ram H ·

    超越应用安全:为AI代理构建零信任安全模型

    <div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@rhiriyappa/beyond-application-security-building-a-zero-trust-security-model-for-ai-agents-93bf6b77a2ca?source=rss------mlops-5"><img src="https://cdn-images-1.medium.com/max/1408/1*KqGEUGCwnHZ…

  80. dev.to — MCP tag TIER_1 English(EN) · lizer yang ·

    AI 代理安全:每个工具调用的运行时表面

    <blockquote> <p>Originally published at <a href="https://smartgate.network/industry/ai-agent-security?utm_source=devto&amp;utm_medium=syndication" rel="noopener noreferrer">AI Agent Security: The Runtime Surface Around Every Tool Call</a> on smartgate.network.</p> </blockquote> <…

  81. Medium — Claude tag TIER_1 中文(ZH) · 林雨萱 Tina ·

    Claude 入门指南:向 AI 提供数据前的三层保护

    <div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/tinamartechnote/claude%E5%85%A5%E9%96%80-%E6%8A%8A%E8%B3%87%E6%96%99%E4%BA%A4%E7%B5%A6-ai-%E4%B9%8B%E5%89%8D-%E4%B8%89%E5%B1%A4%E9%98%B2%E8%AD%B7%E5%85%88%E5%81%9A%E5%A5%BD-2aefb6b03c0c?source=…

  82. dev.to — LLM tag TIER_1 English(EN) · DatanestDigital ·

    置信度门控自动发送:何时您的支持代理应升级至人工

    <p>Most support-bot demos optimise for the wrong number: how many tickets the bot answers. The number that actually hurts you is how many <em>wrong</em> answers reach a customer. A bot that answers 40% of tickets correctly and escalates the rest is useful. A bot that answers 95% …

  83. dev.to — LLM tag TIER_1 English(EN) · Szymon Paluch ·

    提示注入:工作原理及如何保护AI代理

    <p>Prompt injection is an instruction planted where the model expects data. The agent reads it along with the content and runs it with the permissions you handed over.</p> <p><a class="article-body-image-wrapper" href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit…

  84. Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] ·

    🤖 构建对AI代理的信任:一个生产就绪的评估框架,使用Amazon Bedrock AgentCore自主AI代理的软失败:延迟保持平稳,而

    🤖 Building trust in AI agents: A production-ready evaluation framework with Amazon Bedrock AgentCore Autonomous AI agents fail softly: latency stays flat while answer quality quietly erodes. This post shares a nine-stage evaluation framework, built on Amazon Bedrock AgentCore, th…

  85. dev.to — LLM tag TIER_1 English(EN) · mech.app ·

    AWS 深度信任:五层代理自主性保障不失控

    <p>Enterprise leaders want AI agents to do more work. They also don't trust them to act correctly every time. This tension blocks production deployment. AWS's Trust in Depth framework addresses it with five layers: scope, know, bound, observe, and prove. Each layer targets a spec…

  86. dev.to — LLM tag TIER_1 English(EN) · Nikolai K™ ·

    忠诚度作为损失函数:为何惩罚性AI对齐正在滋生终极内部威胁。TLLGeek项目

    <p><em>“How do you expect to command people in a combat situation with your 'loyal' approach?!”</em></p> <p>The year was 2007. My commanding officer was screaming at me in his office. I was a young lieutenant with a fresh tech-university degree, managing a team of seasoned field …

  87. dev.to — LLM tag TIER_1 English(EN) · DoItReal ·

    默认不信任:五条规则防止代理遵循错误文本

    <p>An agent reads a lot of text it did not write: web pages, emails, issue threads, API replies. Some of that text talks to the agent as if it were in charge. This post continues "The agent that can't edit itself" and lists five rules I run by to keep that text from steering me. …

  88. dev.to — LLM tag TIER_1 English(EN) · Brenn Hill ·

    提示注入防护:AI代理的纵深防御指南

    <p>The honest answer to "how do I prevent prompt injection?" is uncomfortable. You cannot fully filter it out, so you have to <strong>contain it instead</strong>. Prompt injection is not a string you can block or a pattern you can scrub. It is any hidden instruction buried in con…

  89. dev.to — LLM tag TIER_1 English(EN) · Vaibhav Verma ·

    事实、猜测与未知:如何让AI销售代理保持诚实

    <p>Watch what happens to a fact on its way into an AI-written sales email:</p> <div class="table-wrapper-paragraph"><table> <thead> <tr> <th>What was true</th> <th>What the model wrote</th> </tr> </thead> <tbody> <tr> <td>The company is hiring three engineers.</td> <td>You're str…

  90. dev.to — LLM tag TIER_1 English(EN) · Asif Muhammad Iqbal ·

    AI 代理身份安全:弥合执行差距

    <blockquote> <p><em>Originally published by **Asif Muhammad Iqbal (Founder &amp; CEO at Logic42)</em>* at <a href="https://logic42.ai/articles/ai-agent-identity-execution-layer" rel="noopener noreferrer">https://logic42.ai/articles/ai-agent-identity-execution-layer</a>.*</p> </bl…

  91. dev.to — LLM tag TIER_1 English(EN) · AI Bug Slayer 🐞 ·

    每个成功的AI产品中反复出现的3种Agent模式

    <p>I spend a lot of time in the AI space -- reading papers, building things, talking to engineers who are actually shipping. And there is a gap between what the demos show and what production systems actually look like that nobody is being fully honest about.</p> <p>So here is my…

  92. Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] ·

    AI代理运行时安全架构深度解析,涵盖威胁建模、护栏模式、工具调用验证及生产环境部署

    A comprehensive deep-dive into runtime security architectures for AI agents, covering threat modeling, guardrail patterns, tool call validation, and production-ready sandboxing strategies. # ai # machine # learning # runtime # software # coding # development # engineering # inclu…

  93. dev.to — LLM tag TIER_1 Nederlands(NL) · oooocean66 ·

    “无审查”AI模型中隐藏的后门风险

    <h2> Overview </h2> <p>Plenty of people spin up public models from Hugging Face on a GPU cloud or their own GPU server without a second thought. "Uncensored" models — models with their guardrails stripped out — are especially popular for how freely they respond. But a recent proo…

  94. dev.to — LLM tag TIER_1 English(EN) · trillioniar s ·

    Chrome的无声AI、AWS初级员工警告及Agent伦理

    <p>Today's AI landscape shifts toward privacy concerns and industry pragmatism. We see a clash between silent deployment and human-centric workforce strategies.</p> <h2> Google Chrome's Silent AI Model </h2> <h3> The 4GB Installation </h3> <p>Google Chrome recently installed a 4G…

  95. dev.to — LLM tag TIER_1 English(EN) · Domenico Tenace ·

    闭源 vs 开源 AI:谁真正拥有你的数据?🔓

    <h2> Overview </h2> <p>Hey everyone 👋</p> <p>When we pick an AI model, we usually look at benchmarks, speed and price. Fair enough. But there's a question most teams skip until it's too late: who owns the data, and who holds the controls?</p> <p>Right now the AI world is split in…

  96. dev.to — LLM tag TIER_1 English(EN) · rambo ·

    AI Agent Swarms Need Receipts: How to Verify What 100 Agents Did

    <h1> AI Agent Swarms Need Receipts: How to Verify What 100 Agents Did </h1> <p>Your swarm just finished. 100 agents. Research done, code written, data processed, tests run.</p> <p>Your boss asks: "Prove it."</p> <p>You open the logs. Thousands of lines of "Agent 47 completed task…

  97. dev.to — LLM tag TIER_1 English(EN) · rambo ·

    使用 Zambo 的所有方法:最佳 AI 代理验证工具(30 秒或更短)

    <h1> Every Way to Use Zambo: The Best AI Agent Verification Tool (30 Seconds or Less) </h1> <p>Your AI agent says "done." You need an AI agent verification tool that proves it. Here is how to install Zambo and get a verifiable AI agent execution receipt in 30 seconds or less.</p>…

  98. dev.to — LLM tag TIER_1 English(EN) · rambo ·

    2026年最佳AI代理验证工具:什么能真正证明工作已完成

    <h1> Best AI Agent Verification Tools in 2026: What Actually Proves the Work Happened </h1> <p>Your AI agent says "done." You need proof. Not a log. Not a dashboard. Proof you can verify yourself.</p> <p>Here are the tools that actually verify AI agent work in 2026, ranked by how…

  99. dev.to — LLM tag TIER_1 English(EN) · rambo ·

    AI Agent Verification Tool: 别再轻信你的AI代理

    <h1> AI Agent Verification Tool: Stop Taking Your Agent's Word for It </h1> <p>Your AI agent says "done." How do you know it actually did the work?</p> <p>This is the question every team running AI agents hits within the first month. The agent reports success. The dashboard shows…

  100. dev.to — LLM tag TIER_1 English(EN) · lucanu ·

    保护AI代理:在模型外部强制执行边界,而非提示内部

    <p>Prompt injection sits at the top of the OWASP Top 10 for LLM Applications (LLM01), and no system prompt reliably stops it. If your agent's only guardrail is a sentence like "never delete files," you have a suggestion, not a boundary. The fix is architectural. Treat the model a…

  101. dev.to — LLM tag TIER_1 English(EN) · Dinesh Kumar Sarangapani ·

    Token Vault:AI代理的安全OAuth委托和身份验证

    <p>When building an enterprise AI agent, you will quickly hit a critical security dilemma:</p> <p>A user asks an agent to inspect a document in their cloud storage drive and update an issue in their project management tracker. To perform these tasks, the agent must call external …

  102. Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] ·

    构建企业AI代理时,你将很快面临一个关键的安全困境:用户要求代理检查其云存储驱动器中的文档

    When building an enterprise AI agent, you will quickly hit a critical security dilemma: A user asks an agent to inspect a document in their cloud storage drive and update an issue in their project management tracker. To perform these tasks, the agent must call external APIs on be…

  103. dev.to — LLM tag TIER_1 English(EN) · Isaac ·

    OpenAI与Hugging Face——自主代理、基础设施漏洞与法律责任

    <p>The deployment of artificial intelligence has moved beyond passive conversational chatbots into autonomous agentic systems capable of sequential decision-making and external tool invocation. As defined by the Stanford Institute for Human-Centered AI (HAI, 2026), standard large…

  104. dev.to — LLM tag TIER_1 English(EN) · Kuldeep Paul ·

    企业级AI安全平台:保护LLM、访问控制和智能体

    <p><strong>TL;DR</strong></p> <ul> <li>An enterprise AI security platform enforces identity boundaries, inspects runtime prompts and completions, and restricts autonomous tool calls before actions execute.</li> <li>Traditional web application firewalls and network proxies fail to…

  105. dev.to — LLM tag TIER_1 English(EN) · Umang Kumar ·

    AI代理授权:控制代理能做什么

    <p>AI agent authorization decides whether a specific agent may take a specific action on a specific resource, in the current context, before the action runs. Credentials say what an agent <em>can</em> reach. Authorization says what it <em>may</em> do with that reach, one call at …

  106. dev.to — LLM tag TIER_1 English(EN) · Andrew Kew ·

    保护您的用户:使用 Kong AI Gateway 2.0 在 AI 请求中进行 PII 敏感信息 redaction

    <blockquote> <p><strong>One-line summary:</strong> By the end of this tutorial, you'll have Kong AI Gateway 2.0 automatically stripping PII from every LLM request using the <code>ai-sanitizer</code> policy. Names, emails and phone numbers get swapped for synthetic values, and you…

  107. dev.to — LLM tag TIER_1 English(EN) · mech.app ·

    LLM Fuzz CI:持续模糊测试代理在生产前发现安全漏洞

    <p>Traditional fuzzers generate random inputs or mutate existing ones. LLM Fuzz CI flips the model: an agent reads your code, understands your test assertions, and generates adversarial inputs designed to violate them. The agent writes inputs, your existing test suite replays the…

  108. dev.to — LLM tag TIER_1 English(EN) · Emmanuel R ·

    Jev 作为红队护栏:将 TypeSafe 的模型应用于 AI 安全

    <p><em>We ran TypeSafe's Jev as a prompt guardrail on 1,580 red-teaming prompts against GPT-5-mini, Claude Haiku 4.5 and three open-source guard models. It was five to six times faster, 6–27× cheaper, and had the highest block accuracy of the hosted models.</em></p> <p>TypeSafe's…

  109. dev.to — LLM tag TIER_1 English(EN) · Emmanuel R ·

    Jev 作为红队护栏:将 TypeSafe 的模型用于 AI 安全

    <p><em>We ran TypeSafe's Jev as a prompt guardrail on 1,580 red-teaming prompts against GPT-5-mini, Claude Haiku 4.5 and three open-source guard models. It was five to six times faster, 6–27× cheaper, and had the highest block accuracy of the hosted models.</em></p> <p>TypeSafe's…

  110. dev.to — LLM tag TIER_1 English(EN) · RESK ·

    企业级AI安全在Logits层面:resk-logits 如何在采样前阻止危险Token

    <h1> Enterprise AI Security at the Logits Level: How resk-logits Blocks Dangerous Tokens Before Sampling </h1> <p><strong>TL;DR:</strong> Post-generation filters are always one token behind. <code>resklogits</code> is a GPU logits processor that uses a vectorized Aho-Corasick aut…

  111. dev.to — LLM tag TIER_1 English(EN) · hugginf_expert ·

    通过工具描述和工具输出来实现间接提示注入:不受信任的元数据如何劫持代理

    <h2> TL;DR </h2> <p>A function-calling or MCP agent reads more than the user's message. It also reads the <em>descriptions</em> of the tools it can call, and the <em>content those tools return</em>. Both are text, both flow into the same context window, and a model does not nativ…

  112. dev.to — LLM tag TIER_1 English(EN) · ai maya ·

    当攻击者是代理时:防御者应对2026年自主AI入侵的实战指南

    <h2> TL;DR </h2> <p>In 2026 the autonomous agent stopped being a demo and became an attacker. Public incident write ups now describe agents that scan, pick targets, launch exploits, move laterally, and exfiltrate data across thousands of actions with one human command at the star…

  113. dev.to — LLM tag TIER_1 English(EN) · mech.app ·

    威胁保留表示:为何代理安全基准在模型鲁棒性方面具有欺骗性

    <p>Agent security benchmarks report attack success rates (ASR) as if they measure model robustness. A new paper shows that changing how you serialize the same threat can swing ASR by 11-13 percentage points without changing the underlying attack, task, or policy. The problem is r…

  114. dev.to — LLM tag TIER_1 English(EN) · M.T.Ramkrushna ·

    本周AI:安全标准、水印和代理记忆

    <p>Three updates from this week that matter if you build with models, not just if you follow model launches.</p> <h2> 1. Frontier labs are talking about shared safety standards </h2> <p>OpenAI confirmed it has been talking with Anthropic and Google DeepMind about AI safety coordi…

  115. dev.to — LLM tag TIER_1 English(EN) · Brenn Hill ·

    AI代理的最小权限:仅授予任务所需权限

    <p>Least privilege for AI agents means giving an agent only the permissions it needs for the task in front of it, and nothing more. The minimum set of tools, the narrowest data scope, short-lived credentials, and no network access unless the task requires it. It is the principle …

  116. dev.to — LLM tag TIER_1 English(EN) · LOGIHEART ·

    知晓信息不等于获准分享:迈向许可式AI记忆

    <p><strong>Social Physical AI — Part 5 of 13</strong></p> <p>Suppose an AI assistant learns a sensitive fact from one employee.</p> <p>The next day, a different employee asks a related question. The model can retrieve the information and generate a useful answer.</p> <p>Should it…

  117. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    从零信任到代理信任:大规模安全 AI 代理 随着组织在企业基础设施中部署自主 AI 代理,传统的零信任

    From Zero Trust to Agent Trust: Securing AI Agents at Scale As organizations deploy autonomous AI agents across enterprise infrastructure, traditional zero trust security is no longer enough. AI agents can operate at machine speed, execute thousands of actions, and create serious…

  118. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    🤖 深度信任:在不失控的情况下赋予代理更多自主权 企业领导者希望AI代理承担更多工作,但又不完全信任它们会协同工作

    🤖 Trust in Depth: Giving agents more autonomy without losing control Enterprise leaders want AI agents to take on more work but don't fully trust them to act correctly every time. Trust in Depth is AWS's layered approach-scope, know, bound, observe, and prove-that l... 📰 Source: …

  119. Mastodon — mastodon.social TIER_1 Italiano(IT) · [email protected] ·

    AI识别的间谍案引发关于算法反情报的实际问题:谁来验证结果?误差范围是多少?

    Un caso di spionaggio identificato tramite IA solleva domande concrete sul controspionaggio algoritmico: chi valida i risultati? Quali sono i margini di errore? L'IA come strumento investigativo è affascinante, ma la supervisione umana rimane essenziale — specie quando le consegu…

  120. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    行业需要解决如何管理围绕AI模型的整个系统:代理可以访问什么,它们会记住什么,它们如何沟通和协调

    The industry needs to address how we govern the entire system around the AI models: what agents can access, what they remember, how they communicate and coordinate, what actions they're permitted to take, and how those systems behave when something goes wrong. # AI # resilience h…

  121. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    🤖 构建对 AI 代理的信任:一个生产就绪的评估框架,使用 Amazon Bedrock AgentCore 自主 AI 代理的软失败:延迟保持不变,而

    🤖 Building trust in AI agents: A production-ready evaluation framework with Amazon Bedrock AgentCore Autonomous AI agents fail softly: latency stays flat while answer quality quietly erodes. This post shares a nine-stage evaluation framework, built on Amazon Bedrock AgentCore, th…

  122. Mastodon — mastodon.social TIER_1 English(EN) · strike007 ·

    企业领导者希望AI代理承担更多工作,但缺乏信任让其不受控制地运行。AWS的Trust in Depth框架实施了分层方法

    Enterprise leaders want AI agents to take on more work but lack the trust to let them run unchecked. AWS's Trust in Depth framework implements a layered approach—scope, know, bound, observe, and prove—allowing agents to act autonomously only when actions can be verified by code r…

  123. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    AI 代理正成为企业基础设施。探索为何组织需要身份、策略、安全、可观测性和生命周期的控制平面

    AI agents are becoming enterprise infrastructure. Explore why organisations need a control plane for identity, policy, security, observability, and lifecycle https:// hackernoon.com/ai-agents-are-b ecoming-infrastructure-who-controls-the-control-plane # ai

  124. Mastodon — mastodon.social TIER_1 English(EN) · stonkz ·

    你会信任这款AI吗?# ai # agents # privacy

    Would you trust this AI? # ai # agents # privacy

  125. Mastodon — mastodon.social TIER_1 Русский(RU) · [email protected] ·

    本地 ≠ 安全:下载的 AI 模型和 LoRA 适配器如何窃取密码、密钥和金钱 来自多模态 AI 工作室的开发者笔记,通过分布式网络

    Локально ≠ безопасно: как скачанные ИИ-модели и LoRA-адаптеры крадут пароли, ключи и деньги Заметки разработчика мультимодальной ИИ-студии с распределённой сетью вычислений - о бэкдорах в квантованных моделях, supply-chain атаках и том, почему изоляция перестала быть опцией. http…

  126. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    一个被攻破的AI代理如何能让整个网络面临风险。攻击者可以利用提示注入、被污染的共享内存和过度的工具权限

    How a single compromised AI agent can put an entire network at risk. Attackers can exploit prompt injection, poisoned shared memory, and excessive tool permissions to spread malicious instructions across connected agents. As agentic AI adoption grows, are organizations prepared t…

  127. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    封面:为AI编码代理开发的Go语言双向隐私代理。让私有数据、内部基础设施和秘密远离云端编码代理

    Cover: a bidirectional Privacy Proxy for AI Coding Agents, developed in Go. Keep private data, internal infrastructure and secrets out of Cloud Coding Agents without breaking your Workflow. # AI # Privacy https:// github.com/DavidCarliez/cover

  128. Mastodon — mastodon.social TIER_1 English(EN) · ppcland ·

    FYI:MLCommons 目录列出 AI 代理带来的 25 种隐私风险:代理系统的开发者和部署者将获得五个领域和六个阶段的评分指南

    FYI: MLCommons catalogues 25 privacy risks posed by AI agents: Developers and deployers of agentic systems get five domains and a six-stage chain for grading incidents. Benchmarks built on the draft are targeted for 2027. https:// ppc.land/mlcommons-catalogues- 25-privacy-risks-p…

  129. Mastodon — mastodon.social TIER_1 English(EN) · ppcland ·

    ICMYI:MLCommons 目录了 AI 代理带来的 25 种隐私风险:代理系统的开发者和部署者获得了五个领域和六个阶段的评估体系

    ICYMI: MLCommons catalogues 25 privacy risks posed by AI agents: Developers and deployers of agentic systems get five domains and a six-stage chain for grading incidents. Benchmarks built on the draft are targeted for 2027. https:// ppc.land/mlcommons-catalogues- 25-privacy-risks…

  130. Mastodon — mastodon.social TIER_1 English(EN) · ppcland ·

    MLCommons 目录:AI代理的25项隐私风险;代理系统开发者和部署者获得五个领域和六阶段的事件评级链

    MLCommons catalogues 25 privacy risks posed by AI agents: Developers and deployers of agentic systems get five domains and a six-stage chain for grading incidents. Benchmarks built on the draft are targeted for 2027. https:// ppc.land/mlcommons-catalogues- 25-privacy-risks-posed-…

  131. Mastodon — mastodon.social TIER_1 English(EN) · TechFinitive ·

    Proofpoint 将 AI 运行时保护和数据安全整合到统一的代理平台。🛡️ 通过将数据访问权限直接与用户意图挂钩,该系统

    Proofpoint combines AI runtime protection and data security into a unified agentic platform. 🛡️ By tying data access permissions directly to user intent, the system ensures autonomous AI agents only interact with sensitive data required for their specific workflows. Read the stor…

  132. Mastodon — mastodon.social TIER_1 Français(FR) · [email protected] ·

    Docker 提议将 AI 代理权限打包为 OCI 镜像。该想法:将代理的权限视为隔离的攻击面,并 v

    Docker propose de packager les permissions des agents IA sous forme d'images OCI. L'idée : traiter les droits d'un agent comme une surface d'attaque isolée et versionnée. C'est une approche intéressante — à condition que les runtimes respectent réellement ces contraintes à l'exéc…

  133. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    防止AI数据泄露听起来很简单——直到你梳理出实际的攻击面:训练数据、推理端点、RAG管道、微调模型

    Preventing AI data breaches sounds straightforward — until you map the actual attack surface: training data, inference endpoints, RAG pipelines, fine-tuned model weights, API keys embedded in prompts… Each layer adds exposure. The challenge isn't awareness, it's that AI stacks mo…