Data covers ERC-8004 agent registrations and Agent Trust Score assessments. Week 33 = week ending Sunday, August 16, 2026.
Week 33 pulled back on both fronts at once. 10,361 new agents registered – down 33.8% from Week 32’s 15,652. Unique owner wallets fell in step, down 35.9% to 9,511, keeping the agents-per-owner ratio essentially flat at 1.09 – still close to the broad, one-agent-per-wallet pattern that returned in Week 32. Median trust score eased 9.2% to 473, staying within the Provisional tier but giving back some of the prior week’s sharp rebound. This week we’re also updating our full trust flag breakdown across the entire indexed agent population, now standing at 387,255 agents.
Week 33 Key Numbers
Week 33 Analysis
Registrations Cool Off After Two Volatile Weeks
After Week 31’s concentration spike and Week 32’s sharp reversal back to broad-based registration, Week 33 simply cooled off on volume. 10,361 new registrations is the lowest weekly total since Week 25’s 5,948, and continues a declining trend that has now run for three consecutive weeks since Week 30’s local peak of 22,989. Unique owners fell by nearly the same proportion as registrations, which is why the agents-per-owner ratio barely moved – this looks like a broad slowdown in registration activity rather than a shift in who is doing the registering.
Trust Score Gives Back Part of the Week 32 Rebound
Median trust score fell from 521 to 473, a 9.2% decline. The score remains inside the Provisional tier (400-599) for a second consecutive week, but the pullback suggests Week 32’s rebound to 521 was not the start of a sustained recovery – trust scores for new registration cohorts continue to swing considerably from week to week, ranging from as low as 200 (Week 31) to as high as 644 (Week 26) over the past eight weeks alone.
Agents-Per-Owner Stays Near One-to-One
At 1.09 agents per owner, Week 33 continues the broad-based registration pattern that returned in Week 32 (1.06), a sharp contrast with Week 31’s 2.54 ratio that signaled fleet-style batch registration. Two consecutive weeks near a 1:1 ratio suggests the concentrated registration event in Week 31 was an isolated episode rather than the start of a new normal – though with only two data points since the reversal, it is too early to call this a stable baseline.
Week-on-Week Comparison: W31 – W33
| Metric | W31 | W32 | W33 | W33 vs W32 |
|---|---|---|---|---|
| New registrations | 21,378 | 15,652 | 10,361 | -33.8% |
| Unique owner wallets | 8,413 | 14,834 | 9,511 | -35.9% |
| Agents per owner | 2.54 | 1.06 | 1.09 | +2.8% |
| Median trust score | 200 | 521 | 473 | -9.2% |
| Trust tier | Elevated Risk | Provisional | Provisional | – |
2026 Weekly Registration History – W27 to W33
| Week | Registered | Unique Owners | Median Trust |
|---|---|---|---|
| W27 | 29,577 | 26,276 | 432 |
| W28 | 24,458 | 21,840 | 464 |
| W29 | 11,026 | 10,339 | 406 |
| W30 | 22,989 | 20,355 | 299 |
| W31 | 21,378 | 8,413 | 200 |
| W32 | 15,652 | 14,834 | 521 |
| W33 | 10,361 | 9,511 | 473 |
Note: Median trust score reflects the median Agent Trust Score of agents registered in that specific week, not the cumulative population median. The trust flag breakdown below covers the full indexed population (387,255 agents) as of Week 33, not just this week’s registrations.
Trust Flag Breakdown: What the Full Agent Population Looks Like
The full indexed agent population now stands at 387,255 agents across eight chains, up from roughly 377,600 last week. Here is the updated trust flag breakdown across that entire population.
| Flag | Agents | Share | What It Means |
|---|---|---|---|
| FARM_DETECTED | 71,215 | 18.39% | Belongs to a Sybil fleet – a single operator controlling multiple agents |
| FEEDER_UNKNOWN | 31,411 | 8.11% | Owner wallet’s funding source is obfuscated or untraceable |
| EIP7702_DELEGATED | 21,180 | 5.47% | Owner has delegated control to a secondary address |
| FEEDER_CEX_VERIFIED | 2,778 | 0.72% | Owner confirmed funded by a centralized exchange – the strongest legitimacy signal available |
| FEEDER_RUG_HISTORY | 760 | 0.20% | Owner wallet was funded by a confirmed rug pull operator |
| CREATOR_RUG_HISTORY | 61 | 0.02% | Agent is controlled directly by a confirmed rug pull creator |
| CREATOR_HONEYPOT_HISTORY | 8 | 0.00% | Owner has previously created a honeypot token contract |
| ERC6551_TBA | 2,156 | 0.56% | Owner is an ERC-6551 Token Bound Account – the real controller is whoever holds the NFT |
| OWNER_MULTISIG | 17 | 0.00% | Owner is a Gnosis Safe multisig, requiring multiple signers to act |
Sybil-farm flagged agents held essentially steady at 18.39% of the population (up marginally from 18.73% last week, as the denominator grew faster than new farm detections) – still by far the largest single flag in the dataset. FEEDER_UNKNOWN ticked up slightly to 8.11%, and ERC6551_TBA-controlled agents grew from 1,942 to 2,156, a proportionally larger jump than most other flags this week, though the category remains a small share of the total population at 0.56%.
The most serious flags remain rare in absolute terms: 760 agents have an owner funded by a confirmed rug pull operator (up 2 from last week), and 61 agents are controlled directly by a wallet with a confirmed rug pull history (unchanged). Confirmed CEX-funded owners – the strongest positive legitimacy signal – crept up to 2,778 agents, still just 0.72% of the population. Honeypot-linked flags and the structural multisig-council flags remain at effectively zero or unchanged from last week.
About This Tracker
ChainAware.ai’s weekly ERC-8004 agent tracker monitors new agent registrations on the ERC-8004 Identity Registry, the number of unique owner wallets behind those registrations, and the median Agent Trust Score of each week’s registration cohort. Agent Trust Score is a 0-1000 score combining owner wallet fraud probability, feeder address analysis, farm detection, and criminal record checks – see the full Agent Trust Score methodology for how the model scores agents and their controlling wallets. Data is updated weekly alongside ChainAware’s PancakeSwap/Uniswap rug pull tracker.