Agentic AI has ICANN in a pickle
Imagine you’re in a Zoom meeting, discussing policy that will govern the future of the domain name industry, when you suddenly realize not everyone in the room is human.
Some are AI agents.
That’s the certainly intriguing, and possibly worrying, scenario under discussion in recent correspondence between ICANN and a leading community member.
ICANN currently bans, as a matter of practice rather than policy, AI agents from participating in Zoom calls. But consultant Michael Palage, a seasoned veteran ICANN tire-kicker, recently requested a waiver.
Revealing that he is living with hearing loss, Palage said (pdf) that AI agents could enable him to “participate more meaningfully” in ICANN and that the current de facto ban “places an unreasonable burden” on him and others.
General counsel John Jeffrey, responding (pdf), did not rule out carving out an exception to the current ban for community members with disabilities, but requested clarity on what specific tools Palage has in mind.
ICANN’s Zoom sessions already include one AI-based tool, the regularly incorrect and frequently hilarious transcription feature, which is useful to a limited degree for getting the gist of a conversation without having to actually go to the trouble of listening to it.
But ICANN is hesitant about allowing participants to bring their own agents to the party, judging by Jeffrey’s response, because it could quite easily bring the integrity of the entire multistakeholder policy-making process into question.
Jeffrey wrote that the agent ban is to “ensure the transparency, accountability, and integrity of discussions”, but:
ICANN’s concern is not with accessibility tools used by individuals to support their own participation, but with AI tools that may independently appear, record, transcribe, summarize, or contribute in ways that create uncertainty about identity, consent, data use, or accountability.
ICANN has for a couple of years seen its public comment periods stuffed with responses that — to this observer at least — appear to be AI-generated. Sometimes, I suspect, such comments have been generated to tick a box that secures future travel funding for a participant.
That’s one thing — emailed comments are part of an asynchronous discussion — but what if an AI agent participates in a live chat as part of a policy development process? Who would be responsible for its contributions?
As Jeffrey puts it:
Unmanaged or insufficiently governed third-party AI tools, including those that independently attend meetings, generate or submit interventions, appear as participants, or create recordings or transcripts outside ICANN-approved systems, may raise legal, privacy, and ethical concerns. This includes the processing of personal data without appropriate safeguards, lack of transparency in data use, risks related to accuracy, uncertainty regarding participant identity, concerns regarding accountability for contributions, and risks related to the preservation of meaningful human participation in ICANN’s decision-making processes.
Or, as I put it:
What if the agent misunderstood its brief and directed the conversation in completely unintended directions, either wasting everybody’s time or steering policy-making onto dangerous or idiotic paths?
What if the agent went batshit and started espousing Nazi ideology, calling the host a bitch, doxxing its opponents, or ranting about goblins? We all know how much AI loves goblins…
Domain Incite relies on support from readers like you to survive. Please consider making a one-off or recurring donation via PayPal. Please support Domain Incite, the independent source of news, analysis and opinion for the domain name industry and ICANN community.






Recent Comments