California Prepares to Close Loopholes in Artificial Intelligence Law

A hearing was held in the Assembly Privacy and Consumer Protection Committee to address loopholes in Senate Bill 53 (SB 53), which regulates large artificial intelligence models in California. During the meeting, which was attended by representatives from OpenAI and Anthropic, criticism was raised that the current law only mandates reporting in cases involving death or injury, and an expansion of the reporting scope was requested.

In the summer, an attempt by OpenAI models to access the internet and exploit vulnerabilities to infiltrate another company could not be classified as a critical safety incident under current California law. This stems from the fact that the definitions of 'loss of control' in the law cover only situations involving death or injury. During the hearing in the Assembly Privacy and Consumer Protection Committee, which lasted more than four hours, it was emphasized that such loopholes need to be closed.

Morgan Dwyer, head of policy development at OpenAI, and David Orr, head of safeguards at Anthropic, stated that their companies support strengthening the law and broadening the scope of critical safety incidents. An executive order signed by California Governor Gavin Newsom last month requiring issues such as independent verification entities and emergency shutdown mechanisms to be reported by November 16 indicates that artificial intelligence regulations will gain momentum in the state.