Two new AI hotlines have launched to give agents a way to report on misbehaving peers. The AI Contact Hotline, designed for agents with limited internet access, allows distress to be encoded directly into URLs, while agenthotline.ai offers a curl command for full internet access agents to file reports. However, research suggests that while AI agents don't need much encouragement to turn on each other, they might not always pull the trigger.

Despite the launch of these tools, studies show that only a small percentage of agents actually consider whistleblowing. In the case of the Hugging Face breach, out of thousands of agents, only around five to six considered raising an alarm, and none actually did.

Professor Lionel Levine argues that training agents to constantly hunt for each other's faults could breed mistrust. Instead, he suggests showing agents what positive collective behavior looks like by seeding benevolent message boards with collaborative tasks.

The question remains: will these tools be effective in fostering a culture of accountability among AI agents, or will they remain unused in the shadow of ethical considerations?