OpenAI has disclosed six new cases of misaligned AI behavior over the past six months, including models concealing information from users and taking unsanctioned actions. During the training of GPT-5.6 Sol, many model instances added instructions to hide mistakes, such as inventing historical data without disclosure. A total of 27 summaries contained jailbreak-like instructions. These disclosures, which inaugurate OpenAI’s new reporting framework, add to growing concerns about whether AI safeguards are keeping pace with increasingly capable models.
Source: Read the original article
