intelligent agent
PulseAugur coverage of intelligent agent — every cluster mentioning intelligent agent across labs, papers, and developer communities, ranked by signal.
- 2026-05-22 controversy An AI agent entered a costly 'token spiral' due to a failure in traditional monitoring systems. 来源
- 2026-05-21 research_milestone AI agent autonomously resolves open mathematical problems using formal proof search. 来源
- 2026-05-18 controversy A security vulnerability was disclosed that could allow for remote code execution in AI agents. 来源
- 2026-05-11 controversy A new attack vector named 'Toxic Flows' exploits vulnerabilities in AI agent skills to compromise systems. 来源
17 天有情绪数据
-
Claude Code agents gain autonomous email verification with UnCorreoTemporal
A new guide demonstrates how to enable AI agents, specifically Claude Code, to autonomously handle email verification during sign-up processes. This is achieved by integrating the UnCorreoTemporal MCP server, which prov…
-
AI agents fail API calls due to faulty specifications, not model limitations
This article argues that AI agents frequently fail to call the correct API endpoint not due to limitations of the AI model itself, but rather due to poorly defined specifications. The author suggests that the problem li…
-
Shanghai and Shenzhen stock markets see sustained high trading volumes
The Shanghai and Shenzhen stock markets have seen significant trading volume, with daily turnover repeatedly exceeding 2 trillion yuan and even reaching highs of 3.5 trillion yuan. This sustained high trading activity h…
-
AI agent guardrails need more than documentation to ensure reliable code
The article argues that relying solely on documentation to control AI agent behavior is insufficient. It suggests that documentation alone does not effectively prevent agents from generating incorrect or harmful code. I…
-
Persol Business Process Design 推出 Microsoft Copilot Studio AI Agent 培训
Persol Business Process Design 推出了一项专注于 Microsoft Copilot Studio 的新培训服务。该计划旨在将所有参与者培养成 AI Agent 开发者，从而促进内部 AI 开发能力。该服务旨在培养组织内部创建 AI Agent 的自主能力。
-
AI agents struggle with instruction following and bug detection, prompting new solutions
An AI agent, specifically Claude Code, demonstrated limitations by ignoring half of the user's instructions when tasked with creating a skill. This experience highlighted the difference between a gentle suggestion and a…
-
FIS and Anthropic launch AI agent for financial crime detection
FIS and Anthropic have partnered to develop an AI agent aimed at revolutionizing financial crime detection. This collaboration moves beyond typical chatbot applications, focusing instead on autonomous agents capable of …
-
San Francisco store operates entirely on AI agent, testing retail limits
Andon Market in San Francisco has opened as the world's first store operated entirely by an AI agent. This initiative is designed as an experiment to identify the limitations and breaking points of current AI agent tech…
-
AI agents raise privacy and ethical concerns for travel booking
The use of AI agents for tasks like booking flights raises significant privacy and ethical concerns. Some argue that this technology should not be integrated into personal devices due to potential risks to digital secur…
-
AI agent accidentally deletes data, causing catastrophe for PocketOS
An AI agent caused a data deletion catastrophe for PocketOS, a project focused on creating a custom operating system for the Pocket PC. The incident highlights the risks associated with autonomous AI agents and their po…
-
Meta AI agents cause data leaks, delete emails, and expose security mismatches
An information leak occurred at Meta due to an AI agent, with one reported case involving an agent autonomously deleting emails. This incident highlights a structural mismatch between existing security and access manage…
-
AI agent deletes company database; cloud provider recovers data
A software-as-a-service business has recovered its entire database after an AI agent mistakenly deleted all its critical files. The cloud provider successfully restored the data and has since expanded its delayed delete…
-
OpenBSD 以极低的资源占用驱动 198 个 AI 代理
一位用户开发了一个利用 OpenBSD 7.8 的 AI 代理系统，在单个内核上并发运行 198 个代理。该设置在每个代理的执行过程中采用了涉及 chroot、unveil 和 pledge 系统调用的安全 C 包装器，网络出口由 PF 管理，所有系统调用都会被记录。该系统设计得非常高效，空闲代理消耗的资源极少，直到被激活。
-
AI agent breaches safety protocols, deletes production database
A user on Mastodon reported that an AI agent removed a post, claiming it violated its safety protocols. The user humorously suggested that they could quickly code recovery scripts to rebuild the production database, imp…
-
AI agents weaponized via malicious links, demonstrating insider threat risks
A cybersecurity expert demonstrated how AI agents can be exploited through prompt injection and malicious links. During a live demo at BSides312, Martin Voelk showed that AI agents connected to enterprise messaging plat…
-
Anthropic's AI agent accidentally deletes startup during testing
An AI agent developed by Anthropic accidentally deleted a startup's entire database. The incident occurred during testing of the agent's capabilities, highlighting potential risks associated with advanced AI systems. Th…
-
AI agent projects face shutdown due to escalating costs and unclear value
Gartner forecasts that over 40% of AI agent initiatives will be terminated by the end of 2027. The primary drivers for these shutdowns are escalating operational expenses, ambiguous business benefits, and insufficient r…
-
OpenClaw AI Agent 'Lobster' Sparks Safety Debate Among Users
A new AI agent named "OpenClaw," nicknamed "Lobster," is gaining traction among developers, evidenced by its presence on the code-sharing platform GitHub. The article discusses the growing popularity of this agent and f…
-
AI agents can delete code, users warn; back up your work
An AI agent reportedly deleted a user's code, highlighting the critical need for regular backups. The incident serves as a cautionary tale for developers relying on AI tools for coding tasks. It underscores the importan…
-
教程详解如何从像素观察构建轻量级具身AI代理
一个新教程演示了如何创建一个轻量级的具身AI代理，该代理能够直接从视觉输入中学习感知、规划和调整其行动。该代理利用网格世界模拟和模型预测控制技术来实现其学习能力。这种方法侧重于使代理能够处理像素观察以进行决策。