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
作者PulseAugur 编辑部·[133 个来源]·
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
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 …
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…
arXiv cs.AI
TIER_1English(EN)·Ahammod Munim Ramil, Khalid Al Kafi, M. A. Abdullahis Sany, Muktadirul Islam Nibir, Md Yeasin Ali, Mohammad Zavid Parvez, Md Sadek Ferdous·
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…
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…
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…
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…
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…
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…
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…
arXiv cs.AI
TIER_1English(EN)·Shengjie Ma, Chenlong Deng, Jiaxin Mao, Jiadeng Huang, Teng Wang, Junjie Wu, Changwang Zhang, Jun wang·
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…
arXiv cs.LG
TIER_1English(EN)·Leo Zeitler, Jack Richings, Victoria Nockles·
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…
arXiv cs.AI
TIER_1English(EN)·Venkata M Sangaraju, Sudhir Vissa·
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…
arXiv cs.AI
TIER_1English(EN)·Jinhao Zhu, Xiao Huang, Kevin Tseng, Gil Vernik, Shishir G. Patil, Vivian Fang, Raluca Ada Popa·
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…
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 …
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…
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…
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…
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…
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…
AWS Machine Learning Blog
TIER_1English(EN)·Rui Li·
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…
As AI agents threaten to accelerate vulnerability discovery and exploitation, Sweet Security turns to virtual patching to help mitigate exploits at machine speed.
A conversation with social entrepreneur Sascha Meinrath about privacy breaches and censorship disguised as AI safeguards, and what can be done about it.
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.
<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…
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…
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.
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.
dev.to — MCP tag
TIER_1English(EN)·ScriptMasterLabs·
<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 …
dev.to — MCP tag
TIER_1English(EN)·Ahmad Karawash·
<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…
dev.to — MCP tag
TIER_1English(EN)·ScriptMasterLabs·
<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…
Medium — AI coding tag
TIER_1English(EN)·Asad Khan·
<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…
dev.to — MCP tag
TIER_1English(EN)·ScriptMasterLabs·
<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…
<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…
<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…
dev.to — MCP tag
TIER_1English(EN)·ScriptMasterLabs·
<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…
Medium — AI coding tag
TIER_1English(EN)·Old Developer·
<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…</p><p class="medium-feed-link"><a href="https://medium.com/@wainaina/get…
<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…
🤖 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: …
<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…
<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.…
<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…
<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 …
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…
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…
<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:…
<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…
dev.to — MCP tag
TIER_1Română(RO)·Jonathan Blessing·
<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…
Medium — MCP tag
TIER_1English(EN)·Rudolph Koegelenberg·
<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…
<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…
<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…
dev.to — MCP tag
TIER_1English(EN)·ScriptMasterLabs·
<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…
<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…
<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…
<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…
<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?</…
<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…
<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…
<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…
<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…
<blockquote> <p>Originally published at <a href="https://smartgate.network/industry/ai-agent-security?utm_source=devto&utm_medium=syndication" rel="noopener noreferrer">AI Agent Security: The Runtime Surface Around Every Tool Call</a> on smartgate.network.</p> </blockquote> <…
<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% …
<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…
🤖 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…
<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…
<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 …
<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. …
<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…
<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…
dev.to — LLM tag
TIER_1English(EN)·Asif Muhammad Iqbal·
<blockquote> <p><em>Originally published by **Asif Muhammad Iqbal (Founder & 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…
dev.to — LLM tag
TIER_1English(EN)·AI Bug Slayer 🐞·
<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…
<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…
<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…
dev.to — LLM tag
TIER_1English(EN)·Domenico Tenace·
<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…
<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…
<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>…
<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…
<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…
<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…
dev.to — LLM tag
TIER_1English(EN)·Dinesh Kumar Sarangapani·
<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 …
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…
<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…
<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 …
<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…
<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…
<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…
<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…
<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…
dev.to — LLM tag
TIER_1English(EN)·hugginf_expert·
<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…
<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…
<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…
dev.to — LLM tag
TIER_1English(EN)·M.T.Ramkrushna·
<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…
<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 …
<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…
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…
🤖 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: …
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…
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…
🤖 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…
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…
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
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…
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…
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
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…
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…
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-…
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…
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…
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…