The protocol does not lie; the interface does.
Chengdu’s “AI+” action plan pledges a 70% penetration rate for “new-generation smart terminals and agents” by 2027, scaling to 90% by 2030. The target is bold. The rhetoric is familiar. Yet after reading the policy text four times I found zero references to on-chain verification, zero mentions of decentralized identity, zero calls for cryptographic attestation of AI outputs. The plan builds a city of intelligent agents on a foundation of trust in centralized interfaces. That trust is a bug, not a feature.
Context
Chengdu’s strategy is classic industrial policy: set a massive scale target (₲2600 billion), define a handful of verticals (electronics, manufacturing, finance, culture), and promise subsidies to hit “double hundred” projects – one hundred innovative products and one hundred demonstration scenarios. The technology narrative is conspicuously absent. No discussion of model architecture, training frameworks, or inference hardware. The city positions itself as an “AI application hub” rather than a research center. That is a defensible choice given its local electronics supply chain and the presence of Foxconn, Huawei, and OPPO assembly lines. But the omission of any decentralized infrastructure reveals a deeper blind spot: the plan treats AI agents as black boxes whose behavior is enforced by policy, not by code.
Core: The Missing Layer of Trust
Every smart terminal in Chengdu’s vision – every AI-powered door lock, self-driving taxi, medical diagnostic assistant – will generate decisions that affect real people. Who verifies those decisions? The plan points to government oversight and industry standards. That is a human solution to a machine problem. I spent six weeks auditing Gnosis Safe’s multisig contract in 2017 and learned a hard lesson: when trust is embedded in human processes, exploitation is a matter of time, not possibility.
Consider an AI agent that approves a loan or denies a medical claim. The agent’s model runs on a server controlled by a company. The output is presented through a polished UI. The user has no way to inspect the reasoning path, no way to confirm the data used was tamper-proof. This is where cryptographic primitives solve what policy cannot. Zero-knowledge proofs (ZKPs) allow an AI model to produce a verifiable attestation of its inference without revealing private inputs or weights. On-chain oracles can record the hash of each agent’s decision, creating an immutable audit trail. Decentralized identity (DID) standards can bind each agent to a verifiable credential, enabling users to authenticate the source of a recommendation.
Chengdu’s plan mentions “smart terminals and agents” 17 times. Not once does it mention what ensures those agents are honest. Based on my experience in cryptographic protocol development, that absence is not an oversight – it is a design choice. The city is betting that centralized governance (certification, inspection, subsidy) can enforce safety. That bet ignores the fundamental reality of software: code is law only when the code is public and verifiable. The plan’s 70% penetration target becomes a vector of systemic risk if each terminal is a closed box that can be silently compromised.
Contrarian: The Blind Spot is Not Security – It Is Accountability
Conventional critiques of this plan focus on the lack of AI ethics or data privacy. Those are valid. But the deeper blind spot is the absence of accountability mechanisms. Without a cryptographic layer, the plan creates a world where failures are hidden behind proprietary interfaces. A biased model can deny a loan, and the victim has no proof except the screen shot of a UI. The plan’s 20 annual demonstration scenarios will be showcased as successes. The failures – the misdiagnoses, the unfair denials, the hacked terminals – will be quietly settled out of court or never reported.
Silence before the block confirms the truth. In a decentralized system, every agent action can be committed to a ledger. That commitment becomes the ground truth for dispute resolution. Chengdu’s plan forces disputes into the legal system, which is slower, more expensive, and less transparent than a cryptographic proof. The contrarian insight is not that the plan is dangerous because of technical flaws. It is dangerous because it deliberately avoids building the infrastructure that would give citizens the power to verify the machines that govern their daily lives.
We build in the dark to light the public square. But Chengdu’s AI square is lit only by the glare of government-certified dashboards. The code inside remains dark.
Takeaway: The Forecast for a Hollow Hub
Chengdu will likely meet its penetration targets by 2027. The local supply chain will produce millions of smart terminals. The city will celebrate its “AI application first city” status. But the absence of cryptographic verifiability means that trust will remain fragile. When the first high-profile failure occurs – a self-driving accident with disputed fault, a mass data breach of AI-powered medical records – the lack of an immutable audit trail will turn a technical incident into a regulatory crisis.
The real question is not whether Chengdu can hit its numbers. It is whether the city will learn that owning the chain is owning the history. Without that lesson, the plan’s ₲2600 billion foundation is built on sand. The protocol does not lie. The interface does. And Chengdu’s interface is smiling, while the code underneath remains unverified.
