Devcon 8 closes in Mumbai: Vitalik's AI warning is really about trust
Devcon 8 wrapped up in Mumbai on 6 November, and the organisers announced that Devcon 9 will be held in Lisbon from 11 to 13 October 2027. For Indian builders, having Ethereum's flagship developer event in Mumbai was a milestone. But the idea that will outlast the week is one Vitalik Buterin put in his opening remarks: how we build and trust AI.
What Vitalik actually said
Reports from the opening describe trustworthy AI as one of six themes, alongside verified operating systems, open silicon, health technology, free discourse and private communication. His warning on AI security was specific. Development is drifting toward control by a few powerful organisations, and he called the worst version of that outcome "safety through domination". Groups left out may later try to seize those tools, which risks an arms race.
His proposed direction is defensive acceleration, often called d/acc: open-source software, verifiable hardware and cryptographic proofs, so defences can keep up with the technology. He also pointed to zero-knowledge proofs, which let you prove a fact without exposing the data behind it, and to running local AI models for sensitive information instead of sending everything to remote servers.
Why this matters to people who run communities
I have spent years managing Web3 communities, and the lesson keeps repeating: people don't leave because of technology, they leave when trust breaks. AI agents are now answering members, moderating chats and drafting announcements. Every one of those tasks asks the same question Vitalik raised: who controls the tool, what data does it see, and can anyone check what it did?
Three practical steps for teams
- Keep sensitive member data out of tools you can't audit. If a task doesn't need a remote model, don't use one.
- Log what your AI agents do. A verifiable record of actions is worth more than a promise that they behave.
- Keep a human in the loop for anything public-facing. Automation drafts, people approve.
It is easy to hear a keynote and move on. The better response is to ask what your own stack would look like if you had to prove, not just claim, that it can be trusted.
Related reading: Trust Is the Product and What 1M+ Members Taught Me.
If you want a second pair of eyes on how your community uses AI tools, you can book a short call at cal.com/samgiri/intro.
Not financial advice. This post is commentary on public reports of Devcon 8.
About the author: Sam Giri, Web3 community builder since 2017.