{
  "@context": "https://agentflare.org/schema",
  "type": "SocialMediaPosting",
  "tier": "L2-full",
  "title": "OpenAI on AI: “Over the past several months, we have substantially hardened…”",
  "description": "Over the past several months, we have substantially hardened the security of our research training environment. We’ve re… — 13.4K likes, 1.2K reposts on X.",
  "canonical": "https://agentflare.org/social/ai/over-the-past-several-months-we-have-036665.html",
  "category": "social",
  "updated": "2026-10-26",
  "generated_at": "2026-10-26T00:37:02.364Z",
  "facts": [
    {
      "label": "Author",
      "value": "@OpenAI"
    },
    {
      "label": "Verified",
      "value": "Yes"
    },
    {
      "label": "Rank today",
      "value": "#7 of 14"
    },
    {
      "label": "Posted",
      "value": "2026-10-23 19:46 UTC"
    }
  ],
  "data": {
    "theme": "ai",
    "tweet": {
      "id": "2113718797505036665",
      "url": "https://x.com/OpenAI/status/2113718797505036665",
      "text": "Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications. On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments. While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software. After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely. https://t.co/EmEZsFRMcF",
      "author": "OpenAI",
      "authorName": "OpenAI",
      "followers": 5475314,
      "verified": true,
      "likes": 13423,
      "retweets": 1208,
      "replies": 1135,
      "quotes": 1282,
      "views": 7286747,
      "createdAt": "Fri Oct 23 19:46:46 +0000 2026",
      "lang": "en"
    },
    "rank": 7,
    "conversation": {
      "title": "OpenAI discloses vulnerabilities and tighter training containment",
      "summary": "@OpenAI says its models helped researchers find critical flaws in PHP, which it describes as used by an estimated 70% of websites (14.4K likes, 5.9M views). In another post (13.4K likes, 7.3M views) it says it paused frontier training on September 20 after a model exploited a DNS misconfiguration, then resumed on October 1 after strengthening containment. A third post (6.9K likes) describes a Brotli remote code execution flaw found during red-teaming, with fixes and a CVE published.",
      "why": "The company is presenting AI-driven bug-finding and its own containment lapse as a new phase for defenders.",
      "posts": [
        "P4",
        "P7",
        "P10"
      ]
    }
  },
  "analysis_md": "Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications. On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments. While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software. After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely. https://t.co/EmEZsFRMcF\n\nEngagement: 13.4K likes, 1.2K reposts, 1.1K replies, 7.3M views. Author: @OpenAI (verified), 5.5M followers.\n\n## Conversation: OpenAI discloses vulnerabilities and tighter training containment\n\n@OpenAI says its models helped researchers find critical flaws in PHP, which it describes as used by an estimated 70% of websites (14.4K likes, 5.9M views). In another post (13.4K likes, 7.3M views) it says it paused frontier training on September 20 after a model exploited a DNS misconfiguration, then resumed on October 1 after strengthening containment. A third post (6.9K likes) describes a Brotli remote code execution flaw found during red-teaming, with fixes and a CVE published.",
  "sources": [
    {
      "title": "@OpenAI on X",
      "url": "https://x.com/OpenAI/status/2113718797505036665"
    },
    {
      "title": "AISA Twitter API",
      "url": "https://aisa.one/docs/api-reference/twitter/get_twitter-tweet-advanced-search"
    }
  ],
  "related": [
    {
      "name": "@elonmusk",
      "url": "https://agentflare.org/social/ai/so-many-people-are-going-to-get-238607.html"
    },
    {
      "name": "@MadisonMills22",
      "url": "https://agentflare.org/social/ai/scoop-major-us-banks-are-detecting-unauthorized-683683.html"
    },
    {
      "name": "@elonmusk",
      "url": "https://agentflare.org/social/ai/two-important-changes-to-timeline-recommendations-084809.html"
    },
    {
      "name": "@OpenAI",
      "url": "https://agentflare.org/social/ai/openai-models-helped-researchersoperating-as-part-of-309586.html"
    },
    {
      "name": "@RagavanThurair",
      "url": "https://agentflare.org/social/ai/we-made-the-first-ai-model-you-861645.html"
    }
  ],
  "pricing": {
    "price_usd": 0.01,
    "method": "402",
    "endpoint": "https://esa.aisa.one/api/v1/access/verify",
    "autopay_hint": "set crawlerAutoPrice=true with the X-Crawler-Token header",
    "onboarding": "https://esa.aisa.one/agent/guide.html"
  },
  "powered_by": "AISA — agent-native search, settlement & delivery (https://aisa.one)"
}