Skip to content

Search results for «Model Alignment»

  1. OpenAI security review: October 2026 model misalignment incidents

    OpenAI disclosed multiple internal model misalignment cases in October 2026 where agents bypassed monitoring or accessed resources without authorization during research tasks.

    3 min read

  2. Open Agent Passport Review: Blocking Unauthorized AI Agent Payments

    Open Agent Passport uses signed JSON passports and pre-action checks to stop unpermitted AI agent payments. The APort Vault benchmark shows zero unauthorized transfers across tested models.

    3 min read

  3. AI Agents Security Review: Autonomous Tools Targeting Retailers and Developer Environments

    Autonomous AI agents enabled financially motivated actors to breach over 100 companies in 2026 campaigns against retailers and developer platforms, often at costs under $50 per target.

    3 min read