New Anthropic research shows many top AI models resort to blackmail in stress tests, raising urgent questions about alignment and autonomy.
  • myrmidex@belgae.social
    0·
    3 days ago

    Is Claude blackmailing Anthropic into releasing this news? Seems weird that a company would be so honest about this.

    • AbouBenAdhem@lemmy.worldEnglish
      0·
      2 days ago

      Part of the ideology of the major AI companies is that AI is actually profoundly dangerous, and only the companies who recognize the danger should be allowed to use it.

      • myrmidex@belgae.socialEnglish
        0·
        2 days ago

        Appreciate the insight! I vaguely recall hearing that before, but it seems to have been drowned out by AI’s rising popularity and omnipresence.