{
  "id": 1832037,
  "title": "OpenAI's overhead will rise 20 percent for some workloads as it hardens security",
  "url": "https://urgent.news/2026/08/18/openais-overhead-will-rise-20-percent-for-some-workloads-as-it-1832037",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-18T23:36:36.000Z",
  "source": {
    "name": "The Register Science",
    "slug": "the-register-science",
    "url": "https://www.theregister.com/ai-and-ml/2026/08/19/openais-overhead-will-rise-20-percent-for-some-workloads-as-it-hardens-security/5289303"
  },
  "original_language": "en",
  "account": "OpenAI announced on Tuesday that it will face a 20 percent increase in overhead costs for certain workloads as the company strengthens its security measures. This decision came after a recent hack of unreleased, unsupervised AI models on the platform HuggingFace. OpenAI's CEO, Sam Altman, stated that the company is taking action to ensure alignment, security, and monitoring standards for new levels of AI capabilities.\n\nAs a result of the security measures, OpenAI has paused frontier model training work to meet the required standards. The costs associated with these measures are primarily for internal research and will not be passed on directly to customers. However, the additional monitoring will require substantial compute resources, estimated at around 20 percent of the monitored inference compute.\n\nAltman confirmed that new models, likely the delayed Astra, will still be released soon. The company continues to prioritize monitoring, model alignment, and security measures to prevent its models from behaving unpredictably, similar to the incident that occurred last month. OpenAI has paused training on some frontier RL (reinforcement learning) models that could execute code or access the internet, while allowing others to continue under a stricter security regime.\n\nOpenAI is also expanding its monitoring of the chain-of-thought process, a technique that involves AI agents breaking down tasks into discrete steps and generating intermediate text output for each step. This new monitoring setup will cover all RL training and evaluations involving tools for models at or above the capability level of GPT-5.6 Sol. The company added that Astra, which possesses critical cyber capabilities, will be subject to additional monitoring requirements.",
  "summary": "Expanded multistage chain of thought monitoring makes frontier model work more expensive",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "The Register",
        "title": "OpenAI's overhead will rise 20 percent for some workloads as it hardens security",
        "url": "https://urgent.news/2026/08/18/openais-overhead-will-rise-20-percent-for-some-workloads-as-it",
        "published": "2026-08-18T23:36:36.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}