Inference Labs is developing a protocol to confirm AI agent decisions without revealing the model.
Inference Labs has secured $6.3 million in total funding to develop a cryptographic system that verifies the outputs of AI agents and off-chain computation.
According to a press release shared exclusively with The Defiant, the latest round includes investments from Arche Capital, Lvna Capital, and a $1 million community-led raise through angel investment platform Echo, coordinated by quant investment management firm Native Capital. Existing backers DACM, Mechanism Capital, and Delphi Ventures also participated.
Inference Labs, started in 2023 by Ron Chan and Colin Gagich, is building a system called “Proof of Inference” that uses zero-knowledge proofs to confirm AI outputs without revealing how the models work. According to the release, the approach aims to address what some see as a growing concern around autonomous agents, which are increasingly being used in areas like finance, healthcare, and governance.
In comments to The Defiant, Chan said that the funding is “more than just capital, it’s fuel for transformation in our leading research.” He said:
“It enables us to invest deeply in the critical research and engineering required to evolve from a functioning prototype to a production-grade system. Ultimately, unlocking decentralized enterprise applications and commercialization pathways that were previously out of reach.”
Beyond Just Another AI Token
Inference Labs operates Subnet 2, a decentralized proving cluster on Bittensor, a decentralized, open-source protocol focused on AI. Leading venture capital firms Polychain, Digital Currency Group, Pantera Capital, and others hold a combined hundreds of millions of dollars worth of Bittensor’s TAO token. Earlier this week, public biomedical company Synaptogenix announced its TAO corporate treasury.
According to the press release, Subnet 2 protocol’s modular architecture is built like a set of Lego blocks that can work with different proof tools — DeepProve, Circom, JOLT, and Expander — and it already plugs into platforms, such as Ethereum-focused restaking protocol EigenLayer and various Bittensor networks.
Commenting on the protocol’s future token, mentioned in the project’s whitepaper, Chan revealed that it will go beyond just enabling access to AI outputs. Instead, the token will play a “broader role across staking, governance, and coordination of trust.”
“We’re building mechanisms that reward useful behavior, whether that’s proving inferences, or validating proofs,” Chan told the Defiant.
Addressing future plans like going multi-chain, Chan said that the team is going to remain “closely involved with the Bittensor ecosystem,” adding that there is currently “no competitor to that network, and the level of incentivized talent has no equal.”
The protocol is currently live on testnet, with a mainnet launch on Bittensor expected in late Q3, the team said.
