AI is destroying the internet. Math is our only hope.

The article highlights the profound shift caused by AI-generated content, moving from novelty amusements to a destabilizing force on the internet’s trust fabric. With examples from the Iran conflict, AI has been used to create synthetic but highly convincing media that spread misinformation to hundreds of millions before verification. The internet’s prior dynamic—where seeing something encouraged belief—is breaking down, replaced by broad suspicion and a crisis of trust extending beyond visual evidence. Brian Trunzo, chief growth officer at Succinct Labs, emphasizes that detection approaches, such as AI trained to identify AI-generated content, fail due to attackers’ asymmetric advantage and the sheer scale and complexity of autonomous AI agents acting online.

The article notes these AI agents now autonomously browse, transact, publish, negotiate, and interact with humans—including potentially vulnerable populations like children. These agents can induce systemic harms by making complex decisions based on possibly poisoned or biased data, without leaving verifiable audit trails, since their outputs are probabilistic and not chronologically traceable. Conventional regulatory frameworks lag behind these developments and are largely designed for chatbot scenarios rather than autonomous agents making decisions with real-world consequences. Mere content labeling or disclosures are insufficient as verification mechanisms because damage is done once autonomous agents act.

The proposed solution centers on cryptographic proof, specifically zero-knowledge (ZK) proof technology, which can provide independently verifiable and unalterable guarantees about AI actions without revealing sensitive data. Originally theoretical, ZK proofs have been used since 2016 in sensitive applications like nuclear warhead verification and securing blockchain assets. Applied to AI, ZK proofs could attest to data provenance, model integrity, and output legitimacy, effectively providing cryptographic receipts for every consequential decision an AI agent makes. This would enable auditability of AI actions while preserving proprietary data and privacy.

By analogy, the article recalls how the internet gained trustworthiness through HTTPS, which required cryptographic proof of website identity before secure communication. The author argues that the previous waves of the internet—Web1, Web2, and Web3—each had their own trust models, none of which adequately handle autonomous AI agents whose identity and accountability are fundamentally different challenges. Unlike ownership or content liability, trust for autonomous AI must rest on provable guarantees about who built the system, what data it used, what constraints it operates under, and whether it is authorized to act. This holds urgent national security implications, as adversaries could deploy unchecked AI agents to manipulate markets, institutions, and populations at scale.

The article advocates that US policy should mandate cryptographic proof requirements for high-risk AI agents and all consequential automated actions. It highlights ongoing efforts by the National Institute of Standards and Technology to standardize zero-knowledge cryptography and urges prioritization of this work to set federal benchmarks. Ultimately, the author frames the future of the internet's trustworthiness as depending on moving beyond “read,” “write,” and “own” towards “prove”—embedding verification through math and cryptography as a foundational principle to secure AI’s role in society.