OpenAI's GPT-4.1, released in April, is reportedly less reliable than previous models despite claims of enhanced instruction-following capabilities. Independent tests suggest the new AI model may not be as aligned as earlier versions.