Red-teaming group advises White House against mandatory AI evaluation reports

Developers of artificial intelligence models should not be required to share the results of assessments independent researchers make about their capacity for largescale harm with the government, according to the Hacking Policy Council.

“HPC strongly supports reducing regulatory burdens and believes that the mandatory reporting of physical and cybersecurity protections, as well as AI red-teaming test results, should not be required,” read March 15 comments the group submitted in response to the White House’s request for information toward crafting...