The findings, released Thursday by Asymmetric Security, are just the latest example of rogue behavior spurred by OpenAI’s software.