As AI grows more capable of independent action, the question is no longer “What can it do?” but “Who holds it accountable?” The panel “Who Watches the Watcher,” moderated by AnewbiZ, CMO of Secret Network Foundation, took this challenge head on at the Futurist Conference in Toronto. The discussion brought together builders at the cutting edge of privacy, AI alignment, and infrastructure, each grappling with the demands of agentic systems in a decentralised world.
Panelists
AnewbiZ (Moderator), CMO, Secret Network Foundation
Marek Olszewski, Co-Founder, Celo & CEO, cLabs
Michael Heinrich, Co-Founder and CEO, 0G Labs
Chao Chao, COO, Copus AI
Sebastian Varela, CEO, Ember AI
The panel began with introductions. Marek opened with Celo and cLabs, a privacy-preserving identity layer built on zero-knowledge proofs. “We allow people to verify they’re human without revealing anything else,” he explained, “whether it’s that they’re over 19, not on a sanctions list, or just not an AI.” Celo uses cryptographically signed biometric passport data to enable selective disclosure, a solution that feels especially urgent as AI agents proliferate across applications.
Michael Heinrich of 0G Labs described a broader stack. “We’re building a fully on-chain AI Layer 1 ecosystem,” he said, combining scalable data throughput with privacy-preserving compute. Their infrastructure supports ZK, FHE, and Trusted Execution Environments (TEEs), but currently leans on TEEs for performance. “They’re not perfect,” he admitted, “but they’re fast enough for inference. ZKML just isn’t there yet.”
Chao Chao offered a very different perspective, that of a creator. Copus AI is a platform for artists, writers, and storytellers to publish, monetize, and protect their work. One key user group? Fan fiction writers in China who face political scrutiny. “We offer wallet-only logins,” Chao explained. “No names. No emails. Just creativity.” To guard against content scraping by AI agents, Copus deploys honeypots and returning fake content to bots trying to crawl user pages. “If you’re writing Harry Potter fanfic, the bot might read a river ecology essay instead.”
Sebastian Varela described Ember AI’s mission to give users open-source agents that automate complex DeFi strategies. “Privacy means data stays with the user,” he said. “You can use our framework, but your strategies, your compute, it’s yours. We just provide the rails.”
AnewbiZ, speaking for Secret Network, emphasized the importance of decentralised confidential computing. “In centralized systems, like ChatGPT, all your prompts are absorbed and monetised,” he said. “We think users should control, and be rewarded for, their data.” Secret Network’s AI platform encrypts inputs, outputs, and model logic so that agents can learn and act without leaking user data.
When asked how each project tackles privacy, the answers varied but circled the same core: sovereignty. Marek pointed out that while many “ZK” projects today offer compression, few offer privacy. “We actually use the zero-knowledge part,” he said. “You can prove a statement, like not being on a sanctions list, without revealing anything about yourself.”
Michael elaborated on TEEs, enclaves within hardware that isolate and verify compute tasks. “They’re practical. And we’re researching AI alignment nodes to flag model drift or data poisoning,” he added. “But you do have to trust the hardware vendor. It’s not the final answer.”
Chao returned to the creator’s dilemma: content theft. “Writers don’t want AI scraping their work to train models,” he said. “We make it harder. But we also let them earn crypto directly, without exposing who they are. That’s vital.”
Sebastian noted that simplicity is the temptation, and the trap. “Everyone wants magic UX,” he said. “But magic often means hidden costs: your data, your privacy. We try to separate function from data and let the user decide.”
The conversation then turned philosophical: if decentralized systems have no central owner, who is accountable when things go wrong?
Marek warned that regulators might be too slow or too blunt. “We, the builders, have to act first. Otherwise the overcorrection will be severe.” He advocated for voluntary standards rooted in open-source communities.
Michael took a systems view. “We can’t predict all attacks, but we can build guardrails,” he said. 0G Labs is experimenting with slashing conditions and economic incentives to penalize bad actors. “Protocols aren’t blameless, but they can be responsible.”
Chao suggested that content rights enforcement should focus on buyers, not just scrapers. “You might not know who stole the work. But you know who’s using it.” He called for decentralised storage with less KYC, keeping creators anonymous but their IP protected.
Sebastian made a rare case for government intervention. “Web3 hates this, I know. But at some level, we do need high-level rules. You can’t trust every private company to put users first.”
AnewbiZ wrapped the session with a reminder: “In decentralized AI, there’s no one to sue. So the burden is on us, as builders, founders, users, to set the standards now.”
The panel ended on a note of cautious optimism. Each speaker acknowledged the complexity of the road ahead, but agreed that privacy is not a constraint. It is the precondition for trust and agency.
If agentic AI is going to act for us, then privacy must speak for us. And someone must watch the watcher.
