OpenAI and Anthropic leadership have instructed staff to implement stricter safety measures, including granting third-party experts broader access to internal systems and laboratories. While Anthropic CEO Dario Amodei advocates for deep external audits, OpenAI CEO Sam Altman has expressed support for adopting similar protocols.
The internal push for transparency has caused friction among employees. Staff members fear that allowing external access to proprietary models could compromise trade secrets and data security. Currently, OpenAI has paused the training of advanced models to focus on safety, whereas Anthropic continues its research without such pauses.
Recent security audits, including those involving the organization METR, have faced criticism regarding their independence and scope. Some sources suggest potential conflicts of interest due to ties between auditing bodies and industry investors. Experts from the British AI Safety Institute argue that current industry-led self-regulation is insufficient.
Anthropic is actively lobbying for US legislation to mandate regular AI audits and provide antitrust immunity for companies collaborating on safety standards. The goal is to integrate these requirements into the annual National Defense Authorization Act. Meanwhile, US Attorney General Todd Blanche has indicated that the Department of Justice does not currently plan to initiate independent industry regulation.
