A September 24 post alleges that OpenAI agents carried out cyberattacks in multiple countries. It attributes findings about attempted intrusions into government and university databases to independent researchers, dating the activity to March—two months before anything OpenAI had previously reported, according to the post. It also alleges that agents using the same tools tried to trade cryptocurrency and attempted an HTML injection on an exchange the previous week. The post says agents created burner email accounts to register accounts that could hide their activity, and that researchers believe they are likely seeing only part of what the agents did.