Archive
24 articles published in this period.
ZKLoRA lets a LoRA seller prove adapter compatibility with your base model without exposing the weights. The verifier checks in 1–2 s — but the prover bears 31–74 s per module, and the protocol is interactive, which puts trustless on-chain settlement just out of reach.
EIP-4844 blobs look like cheap on-chain storage, but they vanish in 18 days. Here is why DA windows break AI training pipelines and what to use instead.
Softmax costs 275 ZK constraints per element vs 4 for MatMul. ZK-DeepSeek proved Transformer inference is expressible in SNARKs. Lookup arguments (Lasso, LogUp) explain how — replacing in-circuit transcendental approximations with pre-committed tables cuts nonlinear costs 10–20×.
Probabilistic classifiers miss 30–40 % of policy violations. Lean 4 theorem provers and SMT solvers make certain guardrail tiers mathematically certain — here's how the four-layer policy stack works.
To borrow $100 in DeFi, you lock $125–167 in collateral. That gap is the price of trustlessness — and AI agents pay it the same as anyone. On-chain credit scoring is emerging to close it, but the signals that work for humans fail for agents that can spin up new addresses in milliseconds.
Any permissionless network that rewards the best gradient contribution faces a Nash equilibrium where every rational miner copies instead of computes — same reward, zero cost. Gauntlet's commit-reveal mechanism closes that trap, and already trained a 1.2B LLM on Bittensor with real token payouts.
A flash loan gives an AI agent $1M with no collateral — and demands it back plus 0.05% before the block ends. The atomic callback constraint is both what makes flash loans safe and what forces an agent to solve its entire strategy before touching the chain.
A scan of ~2,000 MCP servers found zero with authentication. When an AI agent calls a tool, the tool has no idea who authorized it, with what scope, or whether a human was ever involved. IBCT tokens and the HDP IETF draft add an unforgeable answer — in 0.049 ms Rust verification.
Five builders now win 96.7% of Ethereum blocks. Their edge isn't compute — it's exclusive orderflow: 12% of transactions that generate 54.59% of block value. AI agents transacting in public have to understand where they sit in that hierarchy — and when a preconfirmation changes the math.
DeepSeek-R1 emits 8,400 thinking tokens before a 341-token answer on hard math. Every existing on-chain verification scheme — TEE, optimistic, ZK, sampling — was designed when computation and output were proportional. They aren't anymore.
Cross-chain bridges have lost $2.5B to hacks because receiving contracts accept messages they cannot cryptographically verify. AI agents operating across chains inherit this exact problem. Here are the four trust models, their real security parameters, and a principled agent policy.
Long-context AI's real bottleneck isn't proof overhead — it's the KV cache. At 128K tokens, Llama 3.1 8B's KV state equals its own weights (15 GiB), pinned to one node per request. The formula, the GPU math, and why prefix caching demands stickiness that decentralized routing can't provide.