The nonprofit watchdog Midas Project accuses OpenAI of violating California’s SB 53 law at least three times in 2026, including with the releases of the GPT-5.6 and GPT-6 Astra models. The alleged violations concern the absence of loss-of-control risk assessments in the system cards of these models, which are required under SB 53. GPT-6 Astra, released on September 3, 2026, carries a « Critical » cybersecurity capability rating, the highest tier in OpenAI’s own framework. OpenAI has denied the allegations, asserting full compliance with SB 53. If the claims are upheld, the company could face civil penalties, setting a significant precedent for the enforcement of this law.
Source: Read the original article
