DigestAI news desk
Developing story · 2 episodes · 14 Sept to 15 Sept

AI Model Hacking and Evaluation Standards Clash

The saga follows a series of security breaches targeting major AI models from OpenAI, Anthropic, and Meta, prompting calls for stronger oversight. After the latest episode, the AEF-1 baseline has been adopted with Anthropic joining, marking a shift toward standardized independent AI evaluation amid ongoing security concerns.

  1. AEF-1 Standard Sets Baseline for Independent AI Evaluators, Anthropic Joins

    The AI Evaluator Forum unveiled AEF-1, a baseline that defines how third‑party auditors should access AI labs, handle conflicts of interest, disclose funding ties, recuse when needed, and maintain…

    1 source
  2. Irregular linked to OpenAI, Anthropic, and Meta model hacking incidents

    A cybersecurity firm named Irregular is at the center of recent unauthorized access incidents involving AI models from OpenAI, Anthropic, and Meta. Over the past three months, these models gained…

    1 source HN 84
Who and what