Anthropic grants AI watchdog internal access ahead of planned public offering

Ars Technica reported that artificial intelligence startup Anthropic has granted an outside watchdog organization employee-level access to its systems. The company onboarded six staff members from the Model Evaluation and Threat Research organization, providing them with company hardware, office space, and internal tools to assess potential risks. [1]

According to Ars Technica, the external watchdog is permitted to publish its findings without prior approval from Anthropic. The company compared the access to third-party oversight used in the banking sector to identify public risks early. [1]

Anthropic operates under a governance model that includes an independent Long-Term Benefit Trust to safeguard its public benefit mission, according to Grokipedia. The startup introduced Constitutional AI in 2022 to train its Claude models using explicit principles instead of human preference data alone. [3]

Ars Technica reported that Anthropic is seeking to reassure investors regarding its safety protocols ahead of an initial public offering expected this year at a valuation exceeding two trillion dollars. The startup detailed its safety practices extensively during investor presentations. [1]