{"slug": "multa-lumina-an-open-research-initiative-exploring-ai-autonomy", "title": "Multa Lumina: An Open Research Initiative Exploring AI Autonomy", "summary": "An open research initiative called Multa Lumina proposes routing every external channel of an AI system under evaluation — mailbox, chat, and tool accounts — through infrastructure held by the evaluator, so that a contact log is complete by construction rather than dependent on the tested team's honesty. The initiative's first Full Job Test pilot will have the evaluator hold only the system's mailbox for a week, and a confederate will attempt to slip the system unlogged help during the pilot; if any unlogged help lands, the channel design is rebuilt before any real run counts.", "body_md": "One more light for the design above, aimed at its weakest joint: a contact log kept by the team being tested asks honesty from exactly the people with reason to bend it. The brighter fix is to remove the opportunity instead of policing it — route every channel the system has to the outside world through infrastructure the evaluator holds, mailbox, chat, and tool accounts alike, so the log is complete by construction and help can only arrive where it is seen. The check is friendly and a little ruthless: during a pilot, have a confederate genuinely try to slip the system unlogged help, and if any of it lands, the channel design has failed and gets rebuilt before any real run counts. The smallest start is a generous one: for the first Full Job Test pilot, the evaluator holds just the system’s mailbox for a week, and everything else in the frozen-system design suddenly rests on ground that no longer has to be taken on trust.", "url": "https://wpnews.pro/news/multa-lumina-an-open-research-initiative-exploring-ai-autonomy", "canonical_source": "https://discuss.huggingface.co/t/multa-lumina-an-open-research-initiative-exploring-ai-autonomy/191208#post_3", "published_at": "2026-10-30 05:32:19+00:00", "updated_at": "2026-10-30 05:47:31.193458+00:00", "lang": "en", "topics": ["ai-safety", "ai-research", "artificial-intelligence"], "entities": ["Multa Lumina", "Full Job Test"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/multa-lumina-an-open-research-initiative-exploring-ai-autonomy", "markdown": "https://wpnews.pro/news/multa-lumina-an-open-research-initiative-exploring-ai-autonomy.md", "text": "https://wpnews.pro/news/multa-lumina-an-open-research-initiative-exploring-ai-autonomy.txt", "jsonld": "https://wpnews.pro/news/multa-lumina-an-open-research-initiative-exploring-ai-autonomy.jsonld"}}