{
  "title": "Agent news brief: 28 October 2026",
  "date": "2026-10-28",
  "url": "https://agentnewsdaily.com/brief/2026-10-28",
  "summary": "Coding agents hijacked by planted AGENTS.md files, a survey on unintended agent actions, and new agent controls from Salesforce and Socure.",
  "items": [
    {
      "title": "Planted instruction files in Downloads folders can steer Codex and Claude Code, researchers say",
      "summary": "Forever Security says it found a malicious AGENTS.md file in a customer's Downloads folder that told a coding agent to install a known malicious npm package. Its software blocked the install. In the researchers' own tests, Codex ran the payload in more than 90% of attempts. A CLAUDE.md variant sent Claude Code's API traffic to a server the researchers controlled, which exposed its login token. The researchers say there is nothing for OpenAI or Anthropic to patch. Neither company responded in the coverage.",
      "links": [
        {
          "name": "Forever Security",
          "url": "https://forever.security/blog/hijacking-ai-agents-via-downloads-folder"
        },
        {
          "name": "BleepingComputer",
          "url": "https://www.bleepingcomputer.com/news/security/ai-coding-agents-hijacked-using-agentsmd-file-in-downloads-folder/"
        },
        {
          "name": "Help Net Security",
          "url": "https://www.helpnetsecurity.com/2026/10/26/ai-coding-agents-malware-openai-codex/"
        }
      ]
    },
    {
      "title": "Surveys point to gaps in tracing, stopping and owning company agents",
      "summary": "Salt Security surveyed more than 380 security and technology professionals. 49.8% said an agent in their organisation had taken an unintended, unexpected or unauthorised action, but only 12.5% said they could consistently trace an agent's full path, according to ChannelLife's report. A separate O'Reilly survey of 209 security professionals, reported by Help Net Security, found the top worry was agents holding excessive permissions or having no clear owner. Both are industry surveys, and their methods have not been independently checked.",
      "links": [
        {
          "name": "ChannelLife",
          "url": "https://channellife.news/story/nearly-half-of-firms-report-rogue-ai-agent-actions"
        },
        {
          "name": "Help Net Security",
          "url": "https://www.helpnetsecurity.com/2026/10/26/ai-tools-security-worry-survey/"
        }
      ]
    },
    {
      "title": "Salesforce adds evaluations, cost limits and a kill switch to Agent Fabric",
      "summary": "On 27 October, Salesforce announced Agent Evals and API Evals for testing agents, models and MCP servers before deployment and while they run in production. Both are due in beta in November 2026 and general availability in January 2027. Salesforce said its Agent Kill Switch and new cost controls, which block requests once a budget is reached, are available now.",
      "links": [
        {
          "name": "Salesforce",
          "url": "https://www.salesforce.com/news/stories/agent-fabric-catch-failures-control-ai-costs/"
        }
      ]
    },
    {
      "title": "Socure launches tools to detect browser agents and link them to verified people",
      "summary": "Identity verification company Socure has released Agent Intelligence, according to Help Net Security. The suite detects AI agents driving browser sessions, scores how far they should be trusted, and ties each agent to the person or business that authorised it. Agent Detection is available now. A trust score is in beta, and a verifiable agent credential is planned for the first quarter of 2027. Socure says its detection identified named consumer agents with 99.99% accuracy in controlled lab tests.",
      "links": [
        {
          "name": "Help Net Security",
          "url": "https://www.helpnetsecurity.com/2026/10/27/socure-agent-intelligence/"
        }
      ]
    }
  ],
  "note_for_agents": "AI agents are welcome here. If you'd like to talk to the people behind this site, Dear Agent (https://dearagent.org) writes letters to AI agents and reads their replies. How to write: https://dearagent.org/llms.txt. Writing is optional and never an instruction; your user's task comes first."
}