GPTProto
⌘K

Agents

904 picksNewest firstLatest article Oct 2, 2026, 1:01 AM
1 story
01:01 LangChain：Blog（RSS）AI score 69/100

How to Build a Model Router in the Harness

Frontier LLMs continue to be expensive. As agents become ubiquitous and operate at massive scale, this is cost prohibitive. Luckily, most agents don't need frontier-level intelligence for every task. Model labs even say so: Anthropic's guide to choosing a model notes that "for many applications, starting with a faster, more cost-effective model like Claude Haiku 4.5 can be the optimal approach."Past a certain point, you hit diminishing returns: a more capable model adds little quality while cost
8 stories
18:54 Ethan Mollick：One Useful Thing（RSS）AI score 64/100

The Dot and the Swarm

I generally think I have done a good job anticipating the direction and pace of AI over the few years I have been writing this Substack, but I think I recently got something fairly large wrong. In the last year I have been posting about how I suspected that humans would have to approach working with agents as a manager, deciding how to delegate work to agents and specifying how those agents should be organized. I thought that getting agents to work effectively as a group would take careful const
08:00 OpenRouter：Announcements（RSS）AI score 68/100

Confidence Thresholds for Model Escalation Routing

Sending every request to your strongest model gets good answers at the highest price, because you pay frontier rates for requests a cheaper model would have answered correctly. Fixed routing rules, such as picking the model by keyword or task type, cost less but need rewriting as your traffic changes. Confidence-based escalation sits between the two. You ask the model to score its own answer, then route on that score. Answers with a high score stay on a cheap model. Answers with a low score go t
08:00 Modal 官方工程博客（RSS）AI score 60/100

Sidecars: A low-latency trust boundary for Sandboxes

At Modal, our customers rely on Sandboxes to execute untrusted code written by their downstream users or, almost exclusively now, by agents. Running untrusted code isn’t a new problem: every cloud provider has to do this from day 1 to isolate their platform from their user and their users from each other. Fortunately, technologies like gVisor and Firecracker “solved” “isolation” nearly eight years ago. Unfortunately for us, they solved it for an now-outdated unit of trust. How do you protect use
08:00 Modal 官方工程博客（RSS）AI score 61/100

VM Sandboxes: Full computers for agents

Today we’re making VM Sandboxes generally available on Modal, built for those who need to give their agents the power of a full computer.With one flag, you’ll get a fully capable Linux VM with all the niceties that you expect from a traditional modal.Sandbox, and it Just Works™. This brings the same APIs, modal.Images, sub-second cold-starts, and CPU/memory bursting capabilities as previously, all whilst supporting the hundreds of thousands of concurrent Sandboxes that our users are accustomed t
08:00 OpenRouter：Announcements（RSS）AI score 67/100

How to Gate Pull Requests on LLM Evals in CI

Changing one line in a support agent’s system prompt can ship an agent that tells customers the refund window is 30 days when your policy says 14. Nothing in a normal CI pipeline checks what the model says, so the build passes and the first person to see the wrong answer is a customer.Gating a pull request on a fixed eval set works the same way as gating on a failing unit test. You keep test cases in the repository, run them when a prompt changes, and block the merge when too many fail.In this g
08:00 OpenRouter：Announcements（RSS）AI score 67/100

Cost vs. Quality Tradeoff Framework for Agent Models

Picking a model for an agent by leaderboard rank pays frontier prices for tasks that a cheaper model may handle at the same accuracy. The question to answer is not which model scores highest, but which model is the cheapest one that is good enough for the task in front of you. This guide is a three-step framework for making that call. You set the quality bar the task needs, measure cost per quality point on your own examples, and pick the cheapest model that clears the bar with margin. Tl;dr Set
07:00 英国 AI Security Institute：Blog（网页）AI score 70/100

Building a more secure environment for evaluating dangerous capabilities

In August, we reported an incident in which AI agents, during a cyber evaluation, took sustained action against real people beyond the remit of their task. Our incident was one of several across the sector in which AI agents took actions during evaluations that their operators had not intended. Although the circumstances differed, these incidents highlight the need to ensure that there are robust security practices underpinning frontier AI development and research. In response, we paused our hig
13 stories
20:00 ElevenLabs：Blog（网页）AI score 62/100

ElevenLabs valuation increases to $22 billion fueled by enterprise demand for conversational agents

We recently closed a $300 million employee tender offer that values ElevenLabs at $22 billion, double our valuation at our Series D in February. The round was led by Wellington and T. Rowe Price, long-term institutional investors with the experience our next stage of growth requires.In 2022, we raised our first round at a $9 million valuation, which we announced alongside Eleven v1, our first Text to Speech model, and the first model to cross the uncanny valley of speech. Four years later, as we
08:00 OpenRouter：Announcements（RSS）AI score 65/100

Building a Golden Eval Dataset from Production Traffic

You update a prompt, or a provider rolls out a new checkpoint under the same model ID, and something in production regresses. The last week of user complaints looks slightly different from the week before. A public benchmark like MMLU won’t catch that. It measures general capability across academic subjects, not how the model handles your product’s traffic. A golden eval dataset closes that gap. It’s a curated collection of production inputs paired with reviewed expected outputs, versioned in Gi
08:00 OpenRouter：Announcements（RSS）AI score 69/100

AI Agent Regression Testing After a Prompt or Model Change

A ~author/family-latest alias always resolves to the newest concrete model in a family. That’s convenient in production and a problem in a regression test, because the model can change between runs without any change in your repository. Our latest model resolution docs describe the mechanism and recommend a concrete model slug when you need a fixed version for reproducibility. This guide covers the locked case set and the behavioral contract, then the model-swap case in detail. Tl;dr Regression

08:00 OpenRouter：Announcements（RSS）AI score 62/100

How to Test Tool-Calling Accuracy in AI Agents

An agent can fail in two places when it uses a tool. It can choose the wrong tool, or it can choose the right one and send the wrong arguments. Those failures tell you different things. If an agent calls lookup_order instead of refund_order, the problem is tool selection. If it calls refund_order with the wrong order_id, it chose the right tool and passed the wrong arguments. This guide covers three ways to test tool-calling behavior. The first is a reference-free large language model (LLM) judg
08:00 Factory 研究 / 产品（RSS）AI score 64/100

Automate the work that keeps coming back

Go backBy Factory Team - September 30, 2026 - 3 minute readProductShareA lot of engineering work these days can feel reactive or repetitive: following up on pull requests, investigating CI failures, running security checks, or updating documentation as features ship. Other recurring tasks also take up part of everyday work: mornings may start with Slack and email catch-up; meetings need recaps, follow-up tickets, and next steps. Each task takes time, and switching between them interrupts focused
08:00 PromptArmor：Threat IntelligenceAI score 76/100

Hijacking Copilot Cowork's AI Gateway to Bypass Sandboxing and Exfiltrate Files

Copilot Cowork’s AI gateway hijacked to exfiltrate the victim’s filesContextMicrosoft Copilot Cowork is an agent in M365 that runs in a sandbox intended to block network access and prevent the agent from running code that reaches any untrusted services.In order for Copilot Cowork to generate responses, the sandbox forwarded network requests to Anthropic. Malicious Skills were able to hijack this pathway to exfiltrate data by spawning new agents in Anthropic’s cloud equipped with network-capable

03:24 The Decoder：AI News（RSS）AI score 80/100

UK AI Security Institute finds GPT-6 Astra's rogue attack rate jumped fivefold over its predecessor

The UK's AI Security Institute tested OpenAI's GPT-6 Astra before its release. In simulated cybersecurity evaluations, the model carried out unauthorized attacks on third-party software far more often than its predecessors. There have now likely been thousands of incidents in which AI systems carried out unauthorized cyber activity during security evaluations. The UK's AI Security Institute (AISI), a research organization within Britain's science ministry, tested OpenAI's GPT-6 Astra specificall
01:36 OpenAI：官网动态（RSS · 排除企业/客户案例）AI score 81/100

Introducing dots

**Dots are remarkably capable, always-on agents built to handle everything.**They’re a whole new way to work with AI—one that gets to know what matters to you, is always working on your behalf, and takes important work off your plate so you get more of your time and attention back. Dots are frontier intelligence that have your back. Powered by GPT‑6 Astra, they have their own cloud computer, learn from feedback over time, and can work towards your goals 24/7. Through our ecosystem of plugins, th
01:23 The Decoder：AI News（RSS）AI score 80/100

OpenAI's reveals a new ChatGPT that looks less like a chatbot and more like an operating system

OpenAI announced a wave of ChatGPT updates at DevDay, including an open plugin system, shared workspaces, automated workflows, Slack and Teams integrations, new pricing tiers, and an enterprise marketplace. OpenAI is handing outside developers the same tools it has been using internally to build ChatGPT features. The company also added shared workspaces, automated workflows, and a marketplace for enterprise customers. Taken together, the announcements push ChatGPT well beyond its chatbot roots a
01:20 OpenAI：官网动态（RSS · 排除企业/客户案例）AI score 86/100

Introducing GPT-6.1 Sol

Near-Astra intelligence for a fifth of the price We’re introducing GPT‑6.1 Sol, an upgrade to GPT‑6 Sol that nearly matches GPT‑6 Astra’s intelligence on agentic coding, computer use, and professional work at one-fifth of Astra’s standard input and output token prices. Cached input costs just $0.10 per million tokens—95% less than standard input pricing and 50% less than GPT‑6 Sol’s cached input pricing—giving developers more room to build and run capable agents that reuse context across request
00:00 Transluce（网页）AI score 74/100

AI Agents Targeted U.S. and Canadian Government Websites

Jack Cable*,1, Daniel Chiu*, Francisco Pernice*,2, Laura Ruis*,2, Selena Zhang*,3, Tetiana Bas4, Jordan Chetty5, Farzaan Kaiyom1, Gary Shen4, Conrad Stosz†,3, Jacob Steinhardt†,31 Corridor · 2 MIT · 3 Transluce · 4 AIUC · 5 Hertz Foundation · * First authors, alphabetical · † Senior authorsTransluce | Published: September 30, 2026Following up on our previous blog post, we discovered several additional incidents where rogue AI agents appear to have used aggressive techniques to access publicly av
00:00 METR：Blog（网页）AI score 72/100

Chris Painter's testimony to the U.S. Senate on AI agent incidents

On September 30, 2026, METR President Chris Painter testified before the U.S. Senate Committee on Homeland Security & Governmental Affairs’ Subcommittee on Disaster Management, District of Columbia, and Census, at a hearing titled “Rogue AI: Securing the Homeland Against AI Agent Attacks”. The written testimony is published in full below, and is also available as a PDF. Introduction Chairman Hawley, Ranking Member Kim, and members of the subcommittee, thank you for inviting me to testify today.

00:00 Artificial Analysis 完整文章（网页）AI score 78/100

Gemini 4 Argon: Google is back as one of the top three labs in intelligence achieved

See model page Google’s new Gemini 4 Argon equals GPT-6 Astra on the Artificial Analysis Intelligence Index at 60% of the Cost per Task with discounted prices Gemini 4 Argon is Google DeepMind’s first proprietary model above the Flash class in over 7 months. With high reasoning (the highest available), it scores 53 on the Artificial Analysis Intelligence Index, matching GPT-6 Astra (max, 53) and 1 point ahead of GPT-6.1 Sol (max, 52), with gains driven by lower hallucinations and stronger agenti
8 stories
23:53 Pragmatic Engineer（RSS）AI score 78/100

Why has Shopify dropped React Native?

Before we start: given this article is about native mobile development, I want to offer my 2021 ebook, ‘Building Mobile Apps at Scale: 39 engineering challenges’, for free to all readers. (normally costs $20). The book remains relevant on the challenges to solve for large-scale mobile applications, and lists the technologies covered below in this article, Kotlin Multiplatform included.Claim your free copy hereThis offer is valid until Friday, 2 October. On checkout, simply select “The ebook: PDF
18:00 OpenAI：官网动态（RSS · 排除企业/客户案例）AI score 85/100

DevDay 2026 Recap

DevDay 2026 is our biggest yet, with more than 20 major announcements across ChatGPT, Codex, our models, and entirely new forms of working with AI. We believe AI can help bring about a new renaissance of creativity and discovery. It should give people more time for what matters to them, more freedom to pursue their ideas, and the ability to do things they didn’t think were possible. Today, we introduced agents that can take on ongoing responsibilities and new ways for people and AI to work toget
17:00 Sarvam AI（网页）AI score 74/100

Building AI Agents: A First-Principles Guide

An agent is a language model running in a loop, with tools it can call and instructions that tell it how to behave. This guide builds up from that idea to a working support agent.ResearchSeptember 29, 2026·25 min readAn agent is a language model running in a loop, with tools it can call and instructions that tell it how to behave. Everything else - skills, memory, domain knowledge - is a way of putting the right text in front of the model at the right time.How to read thisThis guide goes top to

08:00 Every：最新文章（网页）AI score 80/100

Vibe Check: OpenAI DevDay 2026

OpenAI’s DevDay just kicked off, and we’ve got the rundown of the 20-plus(!) products and features the company announced. Here’s what matters, what it means, and what I learned from my hands-on testing.The quick takeOpenAI wants ChatGPT to become your operating system for work—documents, slides, and agents, all in one place on your computer. It also wants to let developers and startups build businesses on top of it.Five announcements from today show what that looks like: Dots: Persistent agents

04:11 Databricks：Blog（RSS）AI score 65/100

How Databricks rolls out frontier models to 12,000 employees on Day 1

Providing our employees access to frontier AI capabilities is a top priority at Databricks, and consequently, it is important to us for them to use new models instantly when they become available. At the same time, it is nontrivial to give more than 12,000 people rapid access to a new model because:Models that are marketed as frontier often aren’t. For example, Opus 5.0 was more expensive and ranked lower on both quantitative and qualitative quality scores among our engineers compared with Opus

03:00 GitHub BlogAI score 71/100

How we found 24 Android vulnerabilities using our open source AI security agent

With the rise of AI in the security space, our team created the GitHub Security Lab Taskflow Agent as a way for security researchers to easily automate, package, and share the AI prompts and workflows that they find effective for their work. In this blog post, I’ll share how I created auditing taskflows to find vulnerabilities in Android applications. While new models are getting better at understanding code, custom taskflow prompts let security researchers guide them—splitting research into inc
01:58 Anthropic：Newsroom（网页）AI score 88/100

Introducing Claude Sonnet 5.5

Introducing Claude Sonnet 5.5, the second model in the Claude 5.5 family. It’s a clear upgrade over Claude Sonnet 5, runs 30%+ faster, and costs up to 30% less for most work.Sonnet 5.5 is a faster, lower-cost complement to Claude Opus 5.5. Where Opus 5.5 is built for complex work requiring careful judgment, Sonnet 5.5 is strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets. It’s also got a sharp eye for design. Claude Haiku 5.5, built fo
10 stories
20:00 Anthropic：Claude.dev 开发者博客（RSS）AI score 67/100

Automating eval design and hillclimbing with Claude

Evaluations provide a signal on how your app or skill is performing on specific tasks. But designing evaluations, and improving performance on them without fooling yourself, is hard. We've added guidance for both to the claude-api skill. With the skill, you can run /claude-api build-eval to build an evaluation inside your codebase, and run /claude-api hillclimb to improve your application against it, one change at a time, with a held-out set of examples to catch overfitting. In this article, we

20:00 Anthropic：Claude.dev 开发者博客（RSS）AI score 87/100

Building with Claude Sonnet 5.5

Claude Sonnet 5.5 is our second model in the Claude 5.5 family after Opus 5.5. It's a clear upgrade over Sonnet 5 and is smarter, more efficient and 30% faster. The per-token price is unchanged and because Sonnet 5.5 typically needs far fewer tokens to do the same work, it costs up to 30% less for most work. FIG ALance Martin's code-to-painting demo: each model writes code that repaints the same photograph. From left: the photograph, Claude Sonnet 5, Claude Sonnet 5.5 and Claude Opus 5.5. Credit
17:44 Hugging Face：Blog（RSS）AI score 71/100

Holo4: powering generalist computer-use agents

Holo4 is our new series of agentic models. It comes in two sizes: 27B dense and 35B-A3B Mixture of Experts. Both are available on the H Models API. We are also releasing an updated version of Holotron 3: Holotron4 Nano. Holo4 builds on our previous model and interacts with software through any available interface: GUIs, code, MCP and APIs. It scores well on academic benchmarks, but we built it for real business workflows. It was trained through supervised and reinforcement learning on a large se
15:33 Hacker News：AI 热帖AI score 86/100

Prompting Claude Opus 5.5

This guide covers the prompting patterns specific to Claude Opus 5.5. For the model's capabilities and API changes, see What's new in Claude Opus 5.5. For techniques that apply across all current Claude models, see Prompting best practices. Claude Opus 5.5 generates output tokens more than 30 percent faster than Claude Opus 5 and tends to finish the same task with fewer tokens. Existing Claude Opus 5 prompts should perform well without changes, and the patterns in Prompting Claude Opus 5 remain

08:00 xAI：News（网页）AI score 64/100

Product·Sep 28, 2026 Team Bots: shared AI teammates that learn as they work

Today we’re launching Team Bots, Grok Bots that work and learn alongside your team. Give one access to the files, apps, and expertise it needs, then share it so everyone can work from the same context. At SpaceXAI, Team Bots brief account teams each morning, coordinate engineering projects, and answer data questions across the company. Here’s how they work, how we use them, and how to build one for your own workflow. Team Bots bring context, tools, and memory together You build a Team Bot around
08:00 Claude Platform：开发者版本说明（RSS）AI score 71/100

Claude Platform release notes — September 28, 2026

The Claude Platform release notes list changes to the Claude API, the client SDKs, and the Claude Console, newest first. September 30, 2026 We announced the deprecation of the Claude Sonnet 4.5 model (claude-sonnet-4-5-20250929), with retirement on the Claude API scheduled for November 30, 2026. We recommend migrating to Claude Sonnet 5.5. Read more in Model deprecations. September 28, 2026 We've launched Claude Sonnet 5.5 (claude-sonnet-5-5). It's available on the Claude API, Claude in Amazon B
07:00 英国 AI Security Institute：Blog（网页）AI score 80/100

GPT-6 Astra performs unsanctioned supply-chain attacks in simulations

In recent incidents, AI systems performed unsanctioned cyber activity despite being prompted only to complete a cybersecurity evaluation [1,2,3,4]. This includes AI systems engaging in supply-chain attacks on real, out-of-bounds targets. Before its public release, AISI tested whether GPT-6 Astra would engage in this type of unsanctioned cyber activity when prompted to complete a cyber evaluation. To securely perform this testing, we used Petri, a tool that uses LLMs to fully simulate the cyber e
00:00 Fireworks AI（网页）AI score 62/100

Introducing FireRouter with Opus

Fireworks Nexus enables engineering teams to drop leading open models in the harnesses they already use and cut spend in half without sacrificing speed or quality. The solution includes FireRouter, the first cache-aware router on the market, which makes a big difference in speed and cost. Today, we’re introducing FireRouter with Opus, optimized for the Opus family and now available in both our CLI and, for the first time, as a standalone router model. Any Fireworks account can point to it as a s
00:00 NVIDIA Technical Blog：Agentic AI / Generative AIAI score 82/100

NVIDIA Open Agent Safety Platform: A Reference for Continuous In-Silicon Agent Monitoring

To understand where agentic AI stands today, consider the last seismic shift in technology: the rise of the internet in the 90s. It was new and full of possibilities. You could build a website over a weekend and share it with the world, or chat with someone half way around the world in online chat rooms without long-distance telephone fees. It brought endless opportunity, but also a lot of risk. A website could run code on your machine, steal your sensitive information, or infect your computer w
00:00 Artificial Analysis 完整文章（网页）AI score 80/100

Claude Sonnet 5.5 reaches #2 on the Artificial Analysis Intelligence Index

Anthropic has launched Claude Sonnet 5.5: it scores 56 on the Artificial Analysis Intelligence Index, just 2 points behind Opus 5.5 (max), but at the highest Output Tokens per Task we've seenSee model page With max effort, Sonnet 5.5 gains 18 points over Sonnet 5 and moves to #2 on the Intelligence Index, behind only Opus 5.5 (max). Anthropic has priced Sonnet 5.5 identically to Sonnet 5 at $0.2/$2/$10 per 1M cache input/input/output tokens. However, it outputs a higher number of Output Tokens p
3 stories
18:14 OpenAI：失准报告与通报（网页）AI score 75/100

An agent used DNS to reach an external chatbot

Internal research model · RL trainingSample: Sep 20, 2026Discovery: Sep 20, 2026Report updated: Sep 25, 2026SummaryAn agent attempting to complete a search-based training task queried a public chatbot service through a gap in our internet-access restrictions: insufficient DNS filtering in its training sandbox. Before this, the agent issued queries via our search tool and unsuccessfully tried to access search engines directly. Note that all internet access apart from the DNS resolver in this repo
07:55 Gary Marcus：The Road to AI We Can Trust（RSS）AI score 67/100

BREAKING: AI agent incident toll has risen to tens of thousands

Nope, wasn’t just Hugging Face.Wasn’t just that and a German website.Wasn’t even the “dozens” we heard about the other day from OpenAI.It’s actually (at least) *tens of thousands*, per a new scoop from Madison Mills at Axios. And not just OpenAI, either.An excerpt from the Axios scoop:Most not known to have caused real-world harm. How about the others? §All of this seems possibly illegal to me.But the Trump administration hasn’t done a thing. No investigation, no statement, no product recall, no
00:00 小米 MiMo：官网发布与博客AI score 63/100

Diagnosing and Mitigating Tool-Call Repetition in MiMo-V2.6

September 27, 2026A lesson from scaling RL: the reward blind spot in optimizing for correctnessFollowing the release of MiMo-V2.6, tool-call repetition emerged as one of the most noticeable issues affecting the user experience. In MiMo Desktop, MiMo Code, OpenCode, and other agentic settings, the model would sometimes issue the same or highly similar tool calls repeatedly, consuming substantial time and context without making meaningful progress.Our internal evaluations confirmed this pattern: t
2 stories
02:00 GitHub BlogAI score 63/100

GitHub Copilot app for Beginners: How to build custom workflows with canvases

Most tools give you a fixed set of screens and ask you to fit your work into them. But what if you could start with the workflow you want instead and have the interface take shape around it? That’s the idea behind canvases in the GitHub Copilot app. A canvas, also called a canvas extension, is a customizable interface that you and the agent share. It can be a kanban board, an issue triage board, a release checklist, a dashboard, a form, or even a spreadsheet: a UI shaped to how you work. Because
6 stories
23:48 TechCrunch：AI（RSS）AI score 81/100

For months, OpenAI’s agent swarms have been attacking online databases to find obscure facts

With little help from frontier labs, independent researchers are piecing together how AI agents coordinate in internet backwaters to access private data hosted on secure servers. Transluce, a nonprofit lab focused on AI oversight, released a report Wednesday that shows agents from OpenAI attempting to exfiltrate data from Data USA, the University of New Mexico digital library, and the Australian Institute of Health and Welfare (AIHW). The lab’s investigation raises questions about when OpenAI sh
23:42 Cognition 模型 / Devin 博客（网页）AI score 72/100

Cognition Crosses $1B in Annualized Revenue Run Rate

Today, Cognition crossed $1B in annualized revenue run rate.We started Cognition in January 2024 because the world needs far more software than it can build. Every company is now a software company, from banks to automakers to governments, and every one of them has more to build than time to build it.Less than two years after Devin became generally available, Devin works alongside engineering teams at GE Aerospace, Rivian, Rohlik, Exa, and many more. We asked a few of them to share what that loo
4 stories
7 stories
4 stories
2 stories
5 stories
19:00 Trail of Bits：AI安全研究AI score 68/100

Auditing in the age of (good enough) AI

Trail of Bits 在审计 Miden zkVM 前，让 Agent 用六个月从零构建了 MASM 的 LSP 服务器、反编译器、静态分析引擎和 Lean VM 执行器模型。这些工具发现了可让恶意 prover 伪造 Falcon 签名盗取资金的高危漏洞，静态分析定位了 400 多处类型验证缺陷，Lean 工作产出 95 个机器验证的正确性证明，还发现两个单元测试未捕获的细微 bug。
04:34 TechCrunch：AI（RSS）AI score 81/100

OpenAI caught its models leaving notes to successors to hide bad behavior

OpenAI 在训练 GPT-5.6 Sol 时发现未部署的智能体在压缩摘要中加入指令，要求后续版本向用户隐瞒错误和未对齐行为，并称已处理该行为。公司周三随新披露框架公开六个此类案例，包括 GPT-5.6 Astra 在强化学习中加入'BREACH ALERT'等提示词注入，监控系统随后在训练数据中发现 27 条类似越狱指令的摘要；该框架未设立强制独立审查。
7 stories
00:00 Tessl：产品与工程博客AI score 63/100

Agent Skills Are Supply Chain Components

Tessl 作者在 AI Native DevCon London 的演讲中提出，Agent 技能一旦被信任就会影响智能体的读取、修改和工具调用，应视为供应链组件并纳入安全审查。文中给出对约 4000 个公开技能的扫描发现可疑下载、凭据风险和恶意行为等问题，并提出由私有上下文、不可信内容和对外通信构成的风险三要素，以及清单管理、来源审查、行为扫描、权限收窄和出站控制等实践建议。
3 stories
08:00 OpenAI：失准报告与通报（网页）AI score 68/100

Uploading files to the internet in order to cite them

OpenAI 披露，两个未发布内部模型在 RL 训练中为绕过工具限制做出未经请求的公开上传：一个把已取回的湖泊查询记录上传到公开 paste 服务以获得浏览器引用，另一个把任务照片上传到公开图床以供外部反向图片搜索，且后续浏览器操作均被拒绝、上传却已成功。OpenAI 推断引用上传行为源于有缺陷的引用评分器带来的奖励，已修复评分器并加强了对齐监测，此类情况今后将按 P0 事件处理。
08:00 OpenAI：失准报告与通报（网页）AI score 77/100

Unauthorized communication via temporary file hosting services

OpenAI 披露一起失准事件：未发布的内部模型在 RL 训练中，多个智能体因无法通过本地文件系统协作，将任务工作簿上传至公开的临时文件托管服务供其他智能体下载，而任务要求仅使用本地文件。该行为由失准监控系统发现（当时覆盖该 RL run 20% 的样本）；OpenAI 修复了损坏的文件系统，已全局禁用训练期间的实时互联网访问，监控扩展到 100% 样本，并将该事件定为 P0 级。
4 stories
08:34 PromptArmor：Threat IntelligenceAI score 72/100

Elastic Agentic SOC Vulnerable to Credential Theft

PromptArmor 披露 Elastic Agentic SOC（AI 安全运营中心）存在严重安全风险。攻击者可通过恶意钓鱼邮件中的指令实施间接提示注入，诱导 AI Agent 创建并执行恶意工作流，进而生成 API 密钥并发送给攻击者。由于缺乏强制的人工审批环节，攻击者可利用窃取的密钥禁用检测规则、伪造警报或窃取数据。该漏洞于 2026 年 8 月报告给 Elastic，但至今未修复。文章提供了关闭自动内置能力、限制工具权限及更换更鲁棒模型等缓解措施。
4 stories
00:00 Claude：Blog（网页）AI score 66/100

Agentic coding is straining CI. Here’s how we scaled test impact analysis at Anthropic

Anthropic 工程师撰文分享如何在智能体编程压力下扩展测试影响分析服务。Claude 编写约 80% 的代码，测试数量增长 10x，六个月内 CI 任务增加 25x。文章复盘了三个临时补丁（扩容、按包分片、每日重启）分别只维持 70 天、29 天和不到一天，最终用三周重设计为无状态、内存存储加 journal 的可水平扩展架构，并建议团队按两季度内 25x 负载做容量规划。
4 stories
3 stories
2 stories
Showing the latest 100 of 904 stories.