What governments should do on AI safety

Governments should:

  • Set standards to require AI labs to promote safety
  • Require licensing and reporting for very large training runs
  • Security: support labs on information security, physical security, and vetting staff
  • Develop compute governance
  • Strengthen hardware export controls
  • Establish software export controls
  • Fund AI safety research
  • Require and coordinate incident reporting
  • Clarify liability of AI labs

General sources: