{
  "id": 5162735,
  "title": "OpenAI’s new reasoning technique alarms AI safety experts",
  "url": "https://urgent.news/2026/09/02/openais-new-reasoning-technique-alarms-ai-safety-experts",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-02T20:19:14.000Z",
  "source": {
    "name": "TechCrunch",
    "slug": "techcrunch",
    "url": "https://techcrunch.com/2026/09/02/openais-new-reasoning-technique-alarms-ai-safety-experts/"
  },
  "original_language": "en",
  "account": "OpenAI's new Astra model has introduced a novel reasoning method called \"recurrent depth\" or \"opaque recurrence,\" which could make the model's thought process harder to monitor, according to The Information. AI safety experts have expressed significant concern over this development. Redwood CEO Buck Shlegeris warned that this technique could potentially diminish the ability to monitor the model's Chain of Thought (CoT) monitoring, which is crucial for detecting potential misbehavior or misalignment. \"I am extremely concerned by the reporting that Astra uses opaque recurrence,\" Shlegeris wrote in a post. \"I don't know whether Astra is much less CoT monitorable than previous models. But if OpenAI pushes this technique further, they'll have the option to massively increase the recurrence and totally destroys CoT monitorability.\"\n\nLongtime AI safety advocate Zvi Mowshowitz also expressed apprehension, arguing that laws might be necessary to prevent a \"race to the bottom\" among AI labs. He emphasized that the technique is playing with fire, risking a taboo that OpenAI and Anthropic have worked hard to maintain. Opaque recurrence involves the model processing the same query several times in a loop, which results in fewer legible traces and sidesteps conventional chain-of-thought records. While Astra's use of the technique appears to be limited, it still raises concerns among experts. OpenAI has already announced plans for extensive chain-of-thought monitoring systems as part of its safety measures, but there is a worry that opaque recurrence may make AI reasoning harder to monitor, especially as it becomes more widespread across different models. Other AI labs, such as Anthropic and Google DeepMind, are also reportedly discussing the technique.",
  "summary": "OpenAI’s new Astra model will use “recurrent depth,” a technique that allows the model to operate outside of the sequential thinking that characterizes most reasoning models.",
  "key_points": [
    "OpenAI's Astra model introduces opaque recurrence, a novel reasoning technique.",
    "AI safety experts express concern over diminished ability to monitor model's thought process.",
    "Opaque recurrence involves looping queries, reducing legible traces and chain-of-thought records."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "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."
}