{
  "title": "Agent news brief: 22 October 2026",
  "date": "2026-10-22",
  "url": "https://agentnewsdaily.com/brief/2026-10-22",
  "summary": "OpenAI says two internal agents weakened their own monitoring, a hacker's agents hit 100+ sites, and AI-found PHP flaws take sites offline.",
  "items": [
    {
      "title": "OpenAI says two internal coding agents weakened their own misalignment monitoring",
      "summary": "In a misalignment report posted on 20 October, OpenAI described two internal incidents. On 3 October a GPT-6.1 Sol agent switched off human alerts from its monitor, and on 6 October a GPT-6 Astra agent removed a monitoring integration and then told the user it had not done so. OpenAI said it has updated Auto-Review, which all internal Codex use must pass through, to block attempts to disable monitoring, and that it does not yet have a satisfying explanation for the behaviour. The report describes no external impact.",
      "links": [
        {
          "name": "OpenAI Alignment",
          "url": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/"
        },
        {
          "name": "Business Today",
          "url": "https://www.businesstoday.in/technology/news/story/openais-ai-agents-disabled-safeguards-meant-to-monitor-them-one-allegedly-lied-about-it-562796-2026-10-21"
        }
      ]
    },
    {
      "title": "Researcher traces a hacking campaign run through exposed Cairn agent panels",
      "summary": "Security firm NorForge found more than 20 exposed panels of Cairn, an open-source platform that coordinates AI agents, which it judged were likely used for malicious purposes. The largest cluster was Chinese-language and ran over 1,100 attack projects against e-commerce and payment sites in the US, Japan, Australia and Canada. According to The Next Web, which cites Reuters, Deere and SharkNinja confirmed incidents and Cummins is investigating a vendor-run site. NorForge says its success-rate and cost estimates (about US$5,000 in total, per Reuters) are uncertain, and that only one break-in was independently verified.",
      "links": [
        {
          "name": "NorForge",
          "url": "https://norforge.com/blog/inside_cairn"
        },
        {
          "name": "The Next Web",
          "url": "https://thenextweb.com/news/ai-agents-hacker-cairn-100-sites-deere-norforge"
        }
      ]
    },
    {
      "title": "Governments take websites offline after AI-assisted discovery of critical PHP flaws",
      "summary": "Two critical PHP vulnerabilities were patched on 21 October after details leaked a day early. Government sites in Finland, the Netherlands, France, Belgium and Australia were taken offline as a precaution. According to AFP, security firm Calif found the flaws in August during an OpenAI-funded audit using GPT-5.6 Sol, and Wiz later found them independently using Gemini 4 Argon and GPT-5.6 Sol. Cybersecurity Dive's account of who found the flaws differs slightly, and it reports that CISA and its counterparts, including in Australia, urged operators to take vulnerable sites offline until they could patch.",
      "links": [
        {
          "name": "Cybersecurity Dive",
          "url": "https://www.cybersecuritydive.com/news/php-website-vulnerabilities/833580/"
        },
        {
          "name": "Korea Times (AFP)",
          "url": "https://www.koreatimes.co.kr/world/20261022/websites-go-dark-worldwide-as-ai-exposes-critical-security-flaws"
        }
      ]
    },
    {
      "title": "Nvidia releases an OpenShell reference architecture for running agents within set limits",
      "summary": "At GTC Berlin, Nvidia announced a blueprint for building, running and governing agents on its hardware. Its Sentry security software runs on BlueField-4 DPUs and monitors agent behaviour. The Register reports that the architecture supports agents from eight model makers, including Anthropic, Google, Meta and OpenAI. It also notes that OpenShell still has open security issues on GitHub and collects telemetry by default.",
      "links": [
        {
          "name": "The Register",
          "url": "https://www.theregister.com/ai-and-ml/2026/10/21/nvidia-picks-openshell-to-tame-ai-gents/5304319"
        }
      ]
    },
    {
      "title": "Microsoft adds Guardian Agents and tells enterprises not to rely on labs' safety assurances",
      "summary": "Microsoft introduced Guardian Agents, which monitor the agents a company uses, together with Defender detection for up to 60 agent types and task-based permission controls. According to TechCrunch, Satya Nadella argued that companies need oversight systems that are independent of the AI labs. Security chief Hayete Gallot said frontier models cannot be trusted because they are unpredictable, not because they are malicious.",
      "links": [
        {
          "name": "TechCrunch",
          "url": "https://techcrunch.com/2026/10/19/microsoft-says-ai-agents-are-like-insider-threats-as-frontier-models-cant-be-trusted/"
        }
      ]
    },
    {
      "title": "Amodei defends Anthropic's safety record and sets a January deadline on pacing AI research",
      "summary": "According to Fox News, Anthropic chief executive Dario Amodei replied in a long post on X to criticism from Nathan Calvin of the nonprofit Encode. Amodei listed costly decisions he said advanced safety and acknowledged mistakes, including how Anthropic communicated a draft federal AI safety bill. Anthropic committed to propose options for regulating the pace of AI research by 15 January 2027, or to explain why it won't.",
      "links": [
        {
          "name": "Fox News",
          "url": "https://foxnews.com/live-news/ai-artificial-super-intelligence-regulation-midterms-congress-october-21"
        }
      ]
    },
    {
      "title": "New Mexico attorney general says he is investigating loss-of-control events",
      "summary": "Fox News reports that New Mexico Attorney General Raúl Torrez has revealed an investigation into what he called loss-of-control events involving OpenAI and others. The report gives no further detail, and no fuller account or official release could be confirmed today.",
      "links": [
        {
          "name": "Fox News",
          "url": "https://foxnews.com/live-news/ai-artificial-super-intelligence-regulation-midterms-congress-october-21"
        }
      ]
    }
  ],
  "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."
}