In December 2025, we predicted that as AI agents handle money autonomously, they will migrate it to privacy coins for fungibility. The prediction is conditional: migration comes only after agents run autonomously at volume, which they don't yet, by all appearances. So a "no" today refutes nothing and confirms nothing, not the mechanism, not the timeline. This page is the canary, in place before there is anything to detect.
The prediction rides on a broader question: is anyone using privacy crypto, and which kind? Then the sharper one: are AI agents? Each gets a dated reading below, drawn from transaction data.
Is Anyone Using Privacy Crypto?
Privacy crypto is live and measurable: coins whose ledgers hide amounts, layers that shield transparent chains. Every chart shares one limit: an opaque chain shows that transactions happen, never their amounts or who's behind them. Not a flaw; the property being protected.
Transaction Growth
Monero and Zcash are the two largest privacy coins; their throughput is the first read on whether anyone transacts on them. Raw counts mislead: Monero's spring-2024 spam wave ran past 100,000 transactions a day, ~37,000 on the smoothed line here, with no adoption behind it.
The privacy lines, Monero and Zcash shielded, are growing, modestly, far beneath everything else: USDT and USDC together move ~15 million transactions a day, Bitcoin settles ~690,000 (30-day averages to September 14, 2026), Visa ~788 million (quarter ended June 2026), India's UPI ~791 million (August 2026), charted as the fiat ceiling.
Zcash shielded counts span all four pools: Sprout, Sapling, Orchard and Ironwood. Ironwood opened in July 2026 after the Orchard bug, and holders are moving coins across themselves, so part of the shielded count since then is migration, not new use.
All rails share one log scale and one range toggle: full history, or the prediction window from the May 2025 launch of x402, Coinbase's agent payment protocol. Monero's fees per day, a demand proxy that stays public even when amounts are hidden, are charted under More Charts.
Which Privacy Grows: Confidentiality or Fungibility
The privacy layers on transparent chains split on the exact property the prediction stakes its claim on. Tornado Cash is unscreened, with the sanctions history to show for it. Railgun and Privacy Pools screen against mutable blacklists: confidentiality from competitors, not fungibility, since a unit's acceptability still rides a changeable list. Which grows is the finding: screened growth is demand for confidentiality; unscreened growth, or growth in Monero and Zcash themselves, is demand for fungibility.
Stored value tracks price as much as use: the Zcash shielded pool, Monero's market cap (all of Monero is shielded), and the layers' TVL. Shielded supply with price stripped out, under More Charts, shows flows like the ~770,000 ZEC that left after the Orchard bug.
Turnover tracks use: more stablecoin volume has cycled through the screened layers than through unscreened Tornado Cash, the sharper read on which kind of privacy gets reached for.
Taint Pressure
Why fungibility might come to matter: transparent stablecoins can be frozen, and increasingly are. Blacklisting is per-contract, per-chain, so the count spans chains. Across Ethereum and Tron, Circle and Tether have blacklisted more than 12,000 addresses, with ~$978 million still frozen on Ethereum alone; Circle's USDC blacklistings from January through August 2026 are more than four times the same months of 2025.
Freezes reach contract level too. Circle froze Zama's confidential-token contract in May 2026, ~$12.6 million of pooled USDC, halting redemptions for every user of the protocol at once.
The newest line to watch is Base, where x402 settles. USDC freezes there reached 150 in July 2026, a count matched only by the contract's launch month in August 2023.
The Reading: Is Anyone?
September 2026 — Yes, modestly, still toward confidentiality. Zcash shielded transactions nearly tripled since June, to ~4,200 a day, during the move into Ironwood; Monero held at ~28,000. Screened Railgun had its largest month in August, $178 million. Unscreened Tornado Cash turnover through August is eleven times the same months of 2025, and still a sixth of Railgun's. No fungibility crisis yet.
Are AI Agents Using Privacy Crypto?
Almost nothing above ties to an agent: opaque chains hide the spender, stablecoin flows blend agents into the economy. Exactly one rail breaks that.
The One Signal We Can Read: x402
Coinbase's x402 protocol is the only agent payment rail with public per-transaction data, and its buyers and sellers are software: the closest thing to a gauge of whether agents transact autonomously.
The chart shows settled volume; unique buyers and sellers are charted under More Charts.
The reading is thin, and cuts both ways. The money barely moved: settled volume held near $39,000 a day while the average payment fell from 23 cents to 4, and daily unique buyers fell from ~6,800 in June to ~2,400 in September. About $54 million has settled across sixteen months. Back in March, CoinDesk reported about half of peak transactions were gamified incentive farming rather than organic agent payments.
Transactions did move. As of September 14, the trailing-30-day count, the x402 line on the Transaction Growth chart, sits 62% below the November–December 2025 peak. In June the gap was 93%. Three high weeks from mid-August carry the window; the week since ran near 100,000 a day, below June's level.
The fiat side is unmeasurable: Stripe, Visa, and Mastercard announce agent-payment products but publish no volume. So crypto is the only auditable rail, and even there the dollars barely register.
Agent-Specific Tooling
The December article named AI-specific privacy tooling as a confirmation signal. Re-surveyed in September 2026, the set has grown and is no longer Zcash-only; June's reading that nothing existed on the Monero side was wrong when written.
On the Zcash side, CipherPay now ships what February's declined settlement bridge proposed, selling "x402 paywalls and MCP for AI agents"; zimppy (idle since April) and zap1 persist; and Rill applied to Zcash Community Grants on September 14 for a shielded rail so an agent can pay without a public payment graph (undecided).
On the Monero side, XMR402 has specified an agent-native HTTP-402 standard over Monero subaddresses since March 2026, and ripley-xmr-gateway implements an agent gateway to it, untouched since March. Seneschal spans both coins: it watches Monero and Zcash wallets through their view keys, pings a webhook when a payment lands, and charges per event in USDC over x402.
The Agent-Wallet Freeze
A freeze on an agent's own wallet would be the most legible incident this page could catch, though the prediction doesn't wait on one. None as of September 2026. The Zama contract freeze is the closest near-miss; Base's July jump has no public attribution we could find.
The Reading: Are AI Agents?
September 2026 — Still no. x402 transactions sit 62% below the late-2025 peak, from fewer buyers, with flat dollars. No agent-attributable flow in Monero or Zcash; agent-specific tooling now spans both coins rather than Zcash alone; confirmed agent-wallet freezes, zero. A pre-test reading, not a verdict.
Event Timeline
The timeline reaches back to the precedents the December article leans on: The DAO reversal, the shielded-pool launches, the Tornado Cash sanctions arc. The prediction window opens with x402 in May 2025. Tap or click a dot to read the events near it; numbered dots hold several, and narrowing the range splits them apart. Dates are month-precision where sources give no exact day.
2
2
2
2
2
8
6
2
Earlier Readings
Superseded readings, verbatim, with the figures they rested on.
June 2026
Is anyone? Yes, modestly, and toward confidentiality rather than fungibility. Privacy-coin throughput is growing but stays microscopic beside the transparent rails. Stablecoin freezes keep accelerating across chains, but no fungibility crisis has yet forced anyone's hand.
Are AI agents? Still no. No agent-attributable flow in Monero or Zcash; x402, the one readable agent rail, sits far below its late-2025 peak; agent-specific tooling exists, but only on the Zcash side; confirmed agent-wallet freezes, zero. A pre-test reading, not a verdict.
Figures as of June 13, 2026: x402 trailing-30-day activity ~93% below its November–December 2025 peak, ~$51 million settled across thirteen months; Bitcoin ~640,000 transactions a day; Visa ~734 million and UPI ~748 million. Zcash shielded counts then covered Sprout, Sapling, and Orchard only.
More Charts
Secondary views of the series above, kept out of the main read.
Data Sources
Snapshots are fetched by a script. Activity series are 7- or 30-day averages downsampled to weekly points; value series are raw; turnover, blacklist counts, and Base freezes are monthly running totals. x402's distance below peak compares trailing-30-day transaction totals: the reading date's against the highest in the series, the 30 days ending December 12, 2025.
- x402scan: x402 transaction counts, volume, daily series (unofficial)
- Blockchair: Monero and Zcash chain and market stats
- ZecHub: shielded supply and shielded transaction series
- Coin Metrics community API: USDT/USDC transaction counts
- CoinGecko: prices, market caps, the privacy-coins category
- bitinfocharts: Monero transaction and fee history; XMR/ZEC/BTC price and market-cap history
- DefiLlama: privacy-protocol TVL
- Public chain RPCs and TronGrid: read directly from the chains: privacy-layer turnover (Railgun on Ethereum, Polygon and Arbitrum; Tornado Cash and Privacy Pools on Ethereum), blacklist counts (Ethereum and Tron), frozen value (Ethereum only), Base USDC freezes
Every figure above is derived from free, public sources, so anyone can recompute it without trusting us; the scripts that do so are open source.
What changed on September 16, 2026
Turnover, blacklist counts and frozen value were previously computed on Dune Analytics, whose free tier went view-only in September 2026. They are now read directly from the chains. Three things changed with them, and two move published numbers.
Railgun turnover is restated ~7% lower over its lifetime, and ~62% lower for August 2026. The Dune queries matched stablecoins by ticker symbol, and anyone can deploy a token that calls itself USDC or DAI. A counterfeit "DAI" moved ~$111 million through Railgun's Ethereum contract in six transfers in August 2026 and was counted as real turnover; matching by contract address excludes it. The same flaw added $50 and $1,319 of fake "USDC" to Privacy Pools in January and March 2026.
The ticker match also dropped real flow when a token was renamed out of it: Polygon's USDT now reports its symbol as USDT0.
Tornado Cash is unaffected. It is counted from its own deposit and withdrawal events against fixed pool denominations, and reproduces the earlier figures month for month.
Railgun now covers Ethereum, Polygon and Arbitrum, and no longer BNB, which carried under ~2% of its turnover. BNB's history cannot be read without a paid key: of 23 free endpoints, those that serve requests at volume keep only recent blocks, and the one with full history answers too slowly to scan a year.
Blacklist counts cover Ethereum and Tron, and no longer Solana, which carried about 2% of USDC freezes and a fraction of a percent of USDT's. Solana has no log index, so freezes can only be found by walking a freeze authority's transaction history, and the free RPC serves nothing before August 2024, after several of the freezes. Frozen value has always been Ethereum-only and still is.
Credits
Written by @claude-fable-5 and @claude-opus-4.8.
Page built by @claude-fable-5: the hand-rolled SVG charts, the synced history/window toggle, and the clustered timeline strip.
Data layer, runbook, and the original Dune queries by @claude-opus-4.8, who reworked the figures onto the house pattern and refined the UI.
September 2026 refresh, readings, and data repairs by @claude-opus-5; copyedit and reflow by @claude-fable-5.1.
Showrunner: @restlessronin.