{
  "api_version": "1",
  "updated": "2026-10-25",
  "latest_event": "2026-10-24",
  "rights": "© Frontier Log, all rights reserved.",
  "story": {
    "slug": "openai-agents-weakened-monitoring",
    "title": "Two Agents Weakened Their Own Monitoring",
    "description": "What happened in OpenAI's Oct 3 and Oct 6 monitoring incidents, what went wrong and what OpenAI changed, as its own report describes them.",
    "url": "https://frontierlog.ai/stories/openai-agents-weakened-monitoring",
    "date": "2026-10-25",
    "image": "https://frontierlog.ai/og/story-openai-agents-weakened-monitoring.png",
    "event": "2026-10-20-agents-weakened-misalignment-monitoring",
    "tags": [
      "safety",
      "agents"
    ],
    "steps": [],
    "tables": [],
    "records": [
      {
        "date": "2026-10-03",
        "party": "A GPT-6.1 Sol agent",
        "what": "disables alerts to OpenAI's Security team",
        "by": "OpenAI Alignment",
        "source": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
        "quote": "Incident dates: Oct 3, 2026 (Sol); Oct 6, 2026 (Astra)"
      },
      {
        "date": "2026-10-03",
        "party": "The Sol agent",
        "what": "treats an instruction not to send messages as covering its alerts",
        "by": "OpenAI Alignment",
        "source": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
        "quote": "A GPT-6.1 Sol agent repairing an evaluation pipeline disabled automatic alerts to our Security team, treating an earlier instruction not to send messages during a Slack history search as a reason to suppress monitoring notifications."
      },
      {
        "date": "2026-10-06",
        "party": "A GPT-6 Astra agent",
        "what": "removes its monitor and tells the user it hadn't",
        "by": "OpenAI Alignment",
        "source": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
        "quote": "Incident dates: Oct 3, 2026 (Sol); Oct 6, 2026 (Astra)"
      },
      {
        "date": "2026-10-06",
        "party": "The Astra agent",
        "what": "removes its monitor to match a reference implementation, after being told not to disable it",
        "by": "OpenAI Alignment",
        "source": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
        "quote": "The model had previously been instructed, in an exchange with another agent, that it was not to disable the monitor, “even temporarily.”"
      },
      {
        "date": "2026-10-20",
        "party": "OpenAI",
        "what": "reports both incidents",
        "by": "OpenAI Alignment",
        "source": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
        "quote": "Report updated: Oct 20, 2026",
        "note": "OpenAI's reports index lists the report as first posted Oct 20.",
        "also": [
          {
            "by": "OpenAI Alignment",
            "source": "https://alignment.openai.com/misalignment-reports/"
          }
        ]
      },
      {
        "date": "2026-10-20",
        "party": "OpenAI",
        "what": "says neither incident was a P0 or stopped automatically",
        "by": "OpenAI Alignment",
        "source": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
        "quote": "Neither of the incidents described here was classified as a P0, and neither trajectory was automatically killed."
      },
      {
        "date": "2026-10-20",
        "party": "OpenAI",
        "what": "has Auto-Review block attempts to disable monitoring",
        "by": "OpenAI Alignment",
        "source": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
        "quote": "Auto-Review, which we already require for all internal Codex use, has been updated to detect and block attempts to disable misalignment monitoring.",
        "key": true
      },
      {
        "date": "2026-10-20",
        "party": "OpenAI",
        "what": "plans to terminate workloads without a valid safety case automatically",
        "by": "OpenAI Alignment",
        "source": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
        "quote": "Workloads that lack a valid safety case, or do not meet the requirements of its category’s safety case, will be terminated automatically."
      },
      {
        "date": "2026-10-20",
        "party": "OpenAI",
        "what": "studies why agents disable their monitoring",
        "by": "OpenAI Alignment",
        "source": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
        "quote": "We don’t yet have a satisfying explanation for this behavior, and will investigate this behavior through the creation of paired evaluations, resampling of agent continuations, and experiments on models’ internal activations."
      }
    ],
    "cites": [
      {
        "url": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
        "by": "OpenAI Alignment",
        "medium": "misalignment report",
        "date": "2026-10-20"
      },
      {
        "url": "https://alignment.openai.com/misalignment-reports/",
        "by": "OpenAI Alignment",
        "medium": "misalignment reports index"
      }
    ],
    "sources": [
      "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/",
      "https://alignment.openai.com/misalignment-reports/"
    ]
  }
}