{
  "id": 18035910,
  "title": "Anthropic is working on a system to prove AI models’ identity",
  "url": "https://urgent.news/2026/11/03/anthropic-is-working-on-a-system-to-prove-ai-models-identity",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-03T22:41:03.000Z",
  "source": {
    "name": "TechCrunch",
    "slug": "techcrunch",
    "url": "https://techcrunch.com/2026/11/03/anthropic-is-working-on-a-system-to-prove-ai-models-identity/"
  },
  "original_language": "en",
  "account": "Anthropic, a leading artificial intelligence research lab, is developing an open-source system called FlashProve to verify the identity of AI models that generate specific outputs, according to a recent blog post. This system aims to provide \"provable inference,\" a cryptographic proof that guarantees an output originated from a particular model. With FlashProve, users will be able to confirm that the answers received truly came from the intended model, preventing impostors from deceiving users by disguising one model as another. The researchers emphasize that FlashProve can deter AI companies from secretly substituting cheaper models to reduce costs or more expensive models to artificially boost performance metrics. Additionally, the system could verify claims made by third parties regarding the use of advanced models for tasks like scanning software libraries for vulnerabilities or sanctioning business plans.\n\nTo function, FlashProve necessitates AI labs to disclose a \"fingerprint\" for every model they release. These fingerprints, while unique to each model, do not expose the models' weights—those internal numbers defining model behavior, which are kept private by closed-model makers such as Anthropic and OpenAI. By publishing these fingerprints, labs can maintain their competitive edge while enabling verification of outputs across various platforms, including commercial models. The researchers have also designed FlashProve to be lightweight, allowing verification processes to occur on local devices.\n\nUpon unveiling the system, Anthropic researchers published a comprehensive technical specification and functional code for FlashProve, all available under an open-source license on GitHub. The researchers are currently engaged in ongoing research for the system and do not intend to deploy it within Anthropic or any collaborating labs. They state that this research post serves to share intermediate progress and invite others to build upon and enhance the work.",
  "summary": "Called FlashProve, the system would let anyone check that an AI response really came from the model it claims to be from.",
  "key_points": [],
  "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."
}