CoinDesk
Zero-knowledge proofs are the only credible mechanism for verifying autonomous AI agents at scale, according to Brian Trunzo, chief growth officer at Succinct Labs, in a CoinDesk opinion column framed as a manifesto for what he calls Read Write Own Prove.
Trunzo opens with the structural case for why detectors and disclosure regimes have already lost. He points to research showing that simple blur and distortion can drop leading image-detector accuracy to as low as 4%, and to a recent Stanford framing of the core problem as the gap between what AI can do and what society can govern. The pivot is from content provenance to agent provenance: agents that browse, purchase, publish and negotiate have no chronological audit trail, only a single probabilistic pass through opaque parameters.
Why it matters
Trunzo maps the trust crisis onto three prior web epochs. Web1 fixed its identity problem with HTTPS, a cryptographic handshake between browser and server. Web2 scaled on Section 230 liability shields, a bargain that worked for humans posting to timelines and breaks down entirely when the actor is a machine. Web3, in his telling, picked the wrong primitive: tokens optimised for ownership and speculation, not verifiability.
The ZK answer he proposes runs across four layers of the AI stack. At inference, proofs can attest that a specific model with specific parameters produced a specific output. At input, they can verify training data was not poisoned without exposing proprietary datasets. At output, they bind a result to the process that created it. And at identity, they let humans prove they are human and agents prove they are agents, without surrendering privacy.
Market impact
Trunzo frames agent accountability as a national-security question, arguing that foreign adversaries will deploy agents into US markets, institutions and consumer-facing interfaces faster than policy can catch up. He points to NIST's Privacy-Enhancing Cryptography initiative as the technical on-ramp and calls for Congress to require cryptographic proofs of identity and authorisation for any high-risk agent handling financial transactions or interacting with minors. The policy pitch is that liability should attach to the absence of proof, not to the content itself.
The crypto angle is explicit. Trunzo notes that ZK first secured billions in digital assets after moving out of theoretical computer science in 2016, and that the same primitive is now arriving in AI where the bottleneck is no longer compute but truth.
Frequently asked questions
-
Who is Brian Trunzo and what is his argument?
Brian Trunzo is the chief growth officer at Succinct Labs. In a CoinDesk opinion column he argues that AI agents cannot be governed by content labels or disclosure regimes and must instead carry zero-knowledge cryptographic proofs of identity, training data and authorisation.
-
Why do AI image detectors fail, according to the column?
Trunzo cites research showing that simple blur and distortion can drop the accuracy of leading image detectors to as low as 4%. He frames detection as a structurally losing arms race because the attacker holds an asymmetric advantage, the same dynamic that kept antivirus software from eliminating malware.
-
How would zero-knowledge proofs apply to AI agents?
Trunzo lays out four layers: at inference, a proof that a specific model with specific parameters produced a specific output; at input, attestation that training data was not poisoned without exposing proprietary datasets; at output, a cryptographic binding of the result to the process that created it; and at…
-
What policy does the column call for?
Trunzo argues Congress should require cryptographic proofs of identity and authorisation for any high-risk AI agent handling financial transactions or interacting with minors, with liability attaching to the absence of proof rather than to the content itself. He points to NIST's Privacy-Enhancing Cryptography…
-
What is the 'Read Write Own Prove' framing?
It is Trunzo's re-staging of prior web trust fixes. Read came from HTTPS, the cryptographic handshake that made Web1 commerce work. Write came from Section 230, the liability shield that enabled Web2 platforms. Own, the Web3 pitch of user ownership, did not land because tokens optimised for speculation rather than…
Crypto News
CoinTelegraph
WuBlockchain
TheBlock