Brian Trunzo, chief growth officer at Succinct Labs, advocates for zero-knowledge proofs to combat the challenges posed by autonomous AI agents.
By Brian Trunzo|Edited by Cheyenne Ligon Jul 19, 2026, 2:00 p.m. 7 min readMake preferred on ShareShare this articleCopy linkX (Twitter)LinkedInFacebookEmailMake preferred onThe era when AI-generated content was merely a novelty has passed. The recent conflict in Iran illustrated this shift: fabricated videos of American soldiers and Iranian jets circulated widely, gaining belief and reach before verification could occur.
The internet has transformed into a realm of skepticism rather than trust. What was once a reliable source of information is now a breeding ground for doubt, indicating a significant trust crisis that transcends mere appearances.
Brian Trunzo is the chief growth officer at Succinct Labs.
Detection Methods Are Ineffective
The natural instinct might be to enhance detection systems, employing AI to identify AI-generated content. However, this approach has proven ineffective. The leading image detection systems can be easily fooled with minor manipulations, reducing their accuracy to as low as 4%. This ineffectiveness mirrors the struggle antivirus software has faced against malware; attackers typically hold the upper hand.
However, the failure of detection is not the crux of the issue; the problem has already outpaced detection efforts.
AI is evolving beyond content creation; it is now acting. Autonomous agents navigate the internet, conduct transactions, publish material, and engage with humans, sometimes without individuals realizing they are interacting with machines. The consequences of these agents operating at scale can be disastrous.
For instance, an agent trained on misleading data could introduce minor errors in medical billing that escalate into substantial financial losses across a healthcare system. Similarly, a group of commerce agents could exploit pricing gaps, leading to unforeseen losses amounting to billions.
A minor error in training data can lead to significant real-world repercussions.
Once damage occurs, there is no record. An agent's reasoning process is not linear; it involves a complex interaction of numerous parameters, making outputs probabilistic. Asking the same question multiple times can yield varying answers. There is no feasible way to trace the decision-making process back through the constantly shifting variables, nor to audit the agent's training, instructions, or motivations.
A recent Stanford report highlights a pivotal issue: the widening gap between AI capabilities and societal governance. The regulatory landscape is convoluted, with over 90 federal recommendations and a surge of state-level legislation, where more than 1,000 bills were introduced in 2025 alone. However, these frameworks are crafted for chatbots, not for agents that engage in complex transactions.
Content labeling and disclosures offer little help; once an autonomous agent takes action, the harm is done. This presents a verification challenge.
To address this, we need proof.
Proof as a Solution
In this discussion, proof refers to cryptographic, verifiable evidence. It's not merely a claim or a disclosure, but an unalterable guarantee that an AI system has performed as it asserts, using the claimed inputs to produce the stated outputs—without disclosing any underlying data.
This is where zero-knowledge (ZK) proof cryptography comes into play. A ZK proof allows one party to validate a statement's truth without revealing any information beyond that truth. Initially detailed in a 1985 MIT paper The Knowledge Complexity of Interactive Proof Systems, its theoretical elegance gained practical significance in 2016, when it was demonstrated that it could verify nuclear warheads without exposing their designs. Soon after, it found applications in blockchain technology, securing vast sums of digital assets.
Currently, ZK is entering the AI sector, where the primary concern is not computation but truth.
In media, ZK could confirm that a photograph was taken by a legitimate device at a specific time and has remained unaltered—without compromising any sensitive information. This alone could revolutionize the information landscape, but media provenance is just the starting point.
A deeper application lies in AI itself. During inference, ZK can verify that a particular model with designated parameters produced a specific outcome—a verifiable record for every decision made by an agent. At the input stage, it can confirm that training data is untainted, sourced from authorized origins, and compliant with regulations without exposing proprietary datasets. At the output stage, it can cryptographically link a result to its creation process, enabling auditability of significant AI decisions without revealing confidential information. Additionally, ZK allows individuals to authenticate their humanity and agents to confirm their nature without compromising privacy.
A Framework for Restoring Online Trust
In the 1990s, the internet faced a trust crisis. Anyone could launch a server impersonating anyone else. Sensitive information, including passwords and credit card details, traversed the net unencrypted, making secure commerce unfeasible.
The solution was HTTPS. Browsers shifted from trusting websites by default to requiring cryptographic proof: a certificate signed by a recognized authority that links a domain to a public key. Without proof, the browser would not display a padlock icon. Eventually, a lack of proof meant no connection at all. Trust in the web did not stem from promises by platforms but from browsers refusing to transmit sensitive data without verification.
The early web relied on mathematics.
The second iteration, Web2, operated on a different arrangement. Section 230 of the Communications Decency Act granted platforms immunity for user-generated content, resulting in the explosion of the social internet. While this tradeoff suited free speech, it was designed for human interactions rather than autonomous systems. It failed to address the pressing question: who is responsible when the actor is not a person?
Web3 emerged with the promise of user empowerment, enabling individuals to control their data and creators to capture their value. However, this vision faltered, becoming synonymous with NFT speculation, and when valuations plummeted, so did the dream.
Web3’s core instinct was correct: we need a mechanism to replace trust with guarantees. However, ownership alone is insufficient. AI and agents highlight this reality. The critical question is not who owns the platform but whether the actions taken on it can be trusted. Tokens were the wrong foundation; proofs are the appropriate solution.
In an agent-driven internet, both human and machine counterparties require guarantees: who created this system, what data influenced it, what constraints govern it, and whether it is authorized to act. These assurances must hold even when the systems are proprietary, and especially then.
Zero-knowledge cryptography facilitates this by establishing commitments from the outset. A developer can cryptographically mark their training data, allowing ZK proofs to validate identity, provenance, training data, and operational constraints without revealing the underlying information. It’s not about asking for trust; it’s about demanding proof.
This issue extends beyond consumer protection; it touches on national security.
Deepfakes were just the beginning. Agents represent the next phase. Adversarial foreign entities will not only manipulate what Americans perceive—they will deploy agents to influence behavior. These autonomous systems will engage in transactions in our markets, interact with our institutions, and communicate with our children, all without any means for verification of their identity or authorization. This challenge is solvable, but only if the U.S. establishes the necessary frameworks before adversaries exploit the gaps.
Policy must adapt. Congress should mandate that high-risk AI agents, particularly those involved in financial transactions or interactions with minors, possess cryptographic proof of their identity, authorization, and permitted actions, verifiable by any party without disclosing proprietary information. This principle should also apply to the actions of these agents: as they begin transacting on behalf of individuals and businesses at unprecedented speeds and scales, every significant action—be it a payment, contract, trade, or data exchange—should be accompanied by proof of authorization and the constraints under which it operates.
The groundwork for this is already being laid: the U.S. Department of Commerce, through the National Institute of Standards and Technology, is looking into the standardization of zero-knowledge via its Privacy-Enhancing Cryptography initiative. This work should be prioritized, and its outcomes should establish a federal benchmark. Accountability should be linked not to the content produced but to the lack of proof.
HTTPS provided the framework for reading. Section 230 enabled writing. ZK offers a way to prove.
Read Write Own Prove.
Note: The opinions expressed in this piece belong to the author and do not necessarily reflect those of CoinDesk, Inc. or its affiliates.
Latest Crypto News- 1Tether's USDT hits 2-year countdown threatening its position on U.S. crypto platforms1 hour ago
- 2Kraken says simpler options can unlock crypto's next derivatives market1 hour ago
- 3Bitcoin’s quantum problem gets a recovery tool, but not for Satoshi’s 1.1 million coins4 hours ago
- 4Inside Zcash's new node that targets Visa-scale privacy at 50,000 transactions per second8 hours ago
- 5France orders country's internet service providers to block Polymarket16 hours ago
- 6Crypto executives say digital native generations may never need a bank account19 hours ago
- 7DOG Mode explains Bitcoin's next governance fight21 hours ago
- 8Trump targets Brazil's payments system while dollar stablecoins are quietly overtaking country's payments21 hours ago
- 9Here is why a massive $1.6 billion in crypto liquidity is sitting idle and wasting away22 hours ago
- 10Massive bitcoin call spreads target $72,000 by month end, right when the Fed meets23 hours ago
Gate Leads Spot Market Share Gains as CEX Volumes Rise for First Time in Five Months
Gate Leads Spot Market Share Gains as CEX Volumes Rise for First Time in Five Months
CEX trading volumes rose for the first time in five months in June, with spot climbing 15.3% to $1.11T and RWA perpetual volumes surging to a record $311B.
By CoinDesk ResearchJul 13, 2026CEX trading volumes rose for the first time in five months in June, with spot climbing 15.3% to $1.11T and RWA perpetual volumes surging to a record $311B.
Why it matters:
CEX trading volumes rose for the first time in five months in June, with spot climbing 15.3% to $1.11T and RWA perpetual volumes surging to a record $311B.
View Full ReportMore From Opinion