SafetyPage

Lab research on the web · Blog post · Anthropic

Measuring AI capabilities in intelligence targeting and conventional weapons

Read at anthropic.com ↗Read at sourceAll rights reserved

Outline · 72 paragraphs

  1. Models as targeters¶10
  2. Identity correlation and classification¶13
  3. Geolocation from photos¶20
  4. Geolocation from text¶26
  5. Models as weapons developers¶36
  6. Guiding a drone to a target¶40
  7. Dropping a payload on a target¶48
  8. Flying without GPS¶52
  9. Conclusion¶60