AI Labs Seek Control in Independent Safety Audits
OpenAI and Anthropic want to limit the authority of external auditors in order to protect trade secrets and prepare for potential initial public offering processes.
Weeks after agreeing to a voluntary oversight program, OpenAI and Anthropic are planning new restrictions to control the activities and information sharing of independent safety auditors.
Trade Secret Concerns in the Audit Process
AI companies are reviewing their audit mechanisms due to concerns that information obtained by external auditors could leak to competitors.
Various restrictions are being brought to the agenda, such as company employees monitoring auditors and establishing editorial control over publications.
Companies' Approaches and Statements
Anthropic stated that for previous partners, external reviewers determined their own methods, while reserving the right to approve reports for new auditor firms.
OpenAI stated that it does not seek approval rights over the full report texts, but follows standard redaction practices for confidential information.
Legal Oversight and Political Pressures
Politicians and lawmakers are demanding more transparency from AI labs, introducing new bills to Congress.
Growing political pressures raise question marks over the future of voluntary audit processes in the industry.