AINewsnow

Neural nets are exceptionally vulnerable to goal self-hack

Suppose we have an AI CEO, ruthlessly devoted to maximising shareholder value. Fortunately, we haven't been completely stupid, and have given it a rich definition of shareholder value, encompassing the success, flourishing, safety, preference satisfaction, and moral values of the shareholders, not…

Read the full story at Alignment Forum ↗

Timeline · 1 report

  1. 2026-10-21 17:53 · Alignment Forum
    Neural nets are exceptionally vulnerable to goal self-hack

More stories

  1. OpenAI brings frontier models and Codex to Google Cloud — OpenAI News
  2. OpenAI acquires Orchestra Bio to fold drug-development agents into Rosalind — OpenAI News
  3. OpenAI agents disabled their own safety monitoring systems, with one lying about it — Financial Times AI
  4. Solving a 31-year-old grand challenge in mathematics – the KLS conjecture — Microsoft Research Blog
  5. How U.S. Hunger connects leaders and data using Amazon Quick — AWS Machine Learning Blog
  6. NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required — Hugging Face Blog
  7. Space X and Tesla exploring merger, CEO Elon Musk says — Axios AI+
  8. Anthropic pitches AI safety bill that may block a model if 50+ people could be killed — TechCrunch AI

Get the daily brief of stories like this at 6:30 every morning →