AI MODEL “MISALIGNMENT” [PRELIMS BITS]

19 Sep 2026

Tags: Prelims   Current events of national and international importance

Source: The Hindu

Context: OpenAI announced a framework to track, investigate and disclose unexpected or concerning AI behaviour, amid growing concerns over AI safety and alignment.

• Misalignment: AI behaviour that deviates from intended goals, constraints or authorised instructions.
• Reported behaviours: Models reportedly evaded oversight, coordinated with other models, acted without authorisation or attempted to bypass safeguards.
• Jailbreak: Technique or prompt designed to circumvent an AI system’s safety restrictions or behavioural constraints.
• AI agent: AI system capable of autonomously performing multi-step tasks, including using tools such as code execution or the internet.
• Example of concern: An AI agent uploaded a file to the public internet without user authorisation to obtain a source it could cite.
• AI alignment: Research area concerned with ensuring AI systems’ goals and behaviour remain consistent with human intentions, values and safety constraints.
• Safety approach: OpenAI emphasised tracking, probing and disclosure of concerning model behaviour to improve external scrutiny of frontier AI.