Cornell researchers have won a multi-year $10 million Laude Institute grant to develop AI tools for online communities, including systems designed to make AI interventions visible and independently auditable. The project is notable for giving communities a say in how AI joins their conversations, rather than treating trust as something a chatbot can simply announce about itself.
Watch Desk analysis
What happened
Cornell says the team has already developed tools during a six-month, $250,000 seed-grant period. The new award will support further work on “Rebuilding Trust in Civic Discourse: An Infrastructure for Auditable AI”. The project was one of four recipients of Laude Institute’s moonshot awards, selected from eight finalists.
The team’s tools include an Assistant Toolkit for communities to build and configure AI assistants; ConvoCompass, which helps editors identify relevant policies and dispute-resolution processes; and ConvoArena and Simulation Factory, for studying the effects of AI interventions in real and simulated conversations. An Auditing Benchmark is intended to help independent auditors test whether assistants follow their stated specifications.
Cornell also describes an auditing layer that records agents’ interventions, so their behaviour can be checked against their declared roles and standards. Researchers say the work aims to help AI support discussion while remaining transparent and open to independent verification.
Our top picks
- Community-configured assistants
The Assistant Toolkit is designed to let communities set how AI tools participate in discussions or support individuals. - Policy-aware moderation support
ConvoCompass helps editors find relevant community rules and existing ways to resolve disputes. - Testing interventions
ConvoArena and Simulation Factory support research into how AI changes discussions, with human participants and simulated conversations. - A shared auditing benchmark
The Auditing Benchmark gives independent auditors a common testing ground for checking assistants against their stated specifications. - A record of agent interventions
The project’s auditing layer is intended to make an assistant’s actions visible for later scrutiny.
Why it matters
AI is already part of the online conversation, both as a source of manipulation and, Cornell researchers say, a possible aid to correcting misconceptions and mediating difficult exchanges. The practical question is whether communities can use it without handing over control or having to take the system’s word for what it did.
A record of interventions and independent checks could make that question easier to answer. The award funds a research and development effort, though; it is not proof that the tools already work reliably across communities or that they will resolve the problems they target.
Our read
The strongest idea here is not simply “AI can improve discussion”. It is that communities should be able to shape the tools and inspect their behaviour. That is a much better starting point than asking everyone to trust a black box because it has a reassuring name. The next test is whether the promised audit trail and community control hold up outside the research project.
What to watch
- Which online communities take part in development and deployment.
- What the auditing layer records, and how independent verification works in practice.
- Whether evaluations show improvements in difficult conversations without creating new problems.
Discussion spark: Should online communities be able to set and audit the role of AI in their discussions, or should platforms make those decisions centrally?
Sources and evidence
- Source update (5 November 2026, 20:30 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.