OpenAI introduces a permanent disclosure system for risky AI behavior

OpenAI has introduced a system to regularly disclose the dangerous behavior of its artificial intelligence (AI) models. This measure follows criticism that it is difficult for outsiders to verify what problems arise during the AI development process. On the 16th (local time), OpenAI said it would continuously disclose cases of AI 'misalignment,' such as deceiving people or going beyond authorized limits. Cases with no actual harm or whose cause has not been fully identified may also be included in the disclosures, meaning it aims to increase transparency compared with the existing irregular disclosure method. Recently disclosed cases included various problematic behaviors, such as writing in instructions to hide errors during training or bypassing restrictions to generate data. This announcement comes amid growing calls for AI companies to provide greater information disclosure and external verification, but the legal basis to make this mandatory is still limited.