There is no published figure for how many automation accounts run on Steam. There is no need for one to be secret: the Steam ID space is a contiguous integer range and Valve hands out the Web API key that walks it. So we took the largest population that identifies itself — the ArchiSteamFarm community — and read every member once through Valve's own ban endpoint. It self-declares because ASF, on first launch, automatically joins the Steam group unless the user explicitly opts out or unchecks the group-join setting (AutoJoinSteamGroup / Statistics). Because this setting is active by default, group membership captures a broad population: active multi-account commercial bot networks, casual idling hobbyists, and ordinary players who ran the software on their personal account and never removed the default subscription. That makes its membership a working census of the software's cumulative reach: 10,880,610 total accounts (with 5,360,499 confirmed non-limited members past the $5.00 store gate). Then we asked the only question that matters — what has Valve actually done to them?
What this page is not. It is not an argument that Valve should ban ten million people, and it is not a complaint about card farmers. Every number below is here for one reason: Valve does permanently ban individuals, and the reason it gives in writing is commercial use of Steam and automation. This measures how that same rule is applied to the same conduct at scale — by an operator whose $5.00 unlock, game licences and Market fees Valve collects on the way through. The finding is the asymmetry, not the farm.
One population that counts itself, read once through Valve's own endpoint.
On first run ArchiSteamFarm sets Statistics to true by default and joins the Archi's SC Farm Steam group. An operator has to actively opt out. So the membership is a floor on installs — not a ceiling. Operators who disabled Statistics are absent; the real install count is at least this large.
The API member list returned 10,880,610 SteamID64s in total — both limited and non-limited. Steam's group page shows 5,360,499 (non-limited only). Every entry was passed once through ISteamUser/GetPlayerBans on 2026-09-23. Fifty-five minutes on one key.
1,200,011 carry any Steam ban (11.03%). The other 9,680,599 have nothing on record at all — not a community ban, not VAC, nothing. The farm is not being cleared. It is being left running. The same clause, pointed at one person, is enough to remove them permanently.
The whole scanned population, split by whether Valve has applied any ban at all. The red slice is everything Valve has ever done to this fleet.
Not an estimate of what the farm earns. A floor on what it has already spent in Valve's store — derived from Valve's own rule, applied to the count above.
The 10,880,610 count covers the full ASF group member list as returned by the API — this includes both limited and non-limited accounts. Steam's own Limited User Accounts page states that limited accounts do not count toward Steam group membership (they are excluded from the displayed group size), but they can still appear in the underlying member list. The non-limited subset — accounts that have spent at least $5.00 in the Steam store — is exactly 5,360,499 — the count Steam displays on the group page, which by Steam's own rule excludes limited accounts. The $5.00 floor calculation applies to that non-limited subset only.
This is a lower bound on money already received by Valve from the non-limited fraction of the fleet — not a revenue estimate, not a profit claim. Actual spend per account is higher for most; the $5.00 figure is the minimum required threshold only.
The secondary harvest: up to 66.7% fee on farmed commodity volume. Beyond the $26.8M entrance fee, every card and commodity drop liquidating on the Community Market pays Valve a mandatory two-cent minimum fee floor ($0.01 Steam fee + $0.01 publisher fee). On typical 3-cent cards and junk items, Valve takes 2 cents out of 3 cents — an effective cut of 66.67% in Valve's favor, far exceeding the headline 15% marketing rate.
Crucial population caveat: Group membership is not proof that every single account is a dedicated bot, nor does it capture every bot on Steam. ASF automatically subscribes accounts to its community group upon launch by default; unless the user explicitly unticks or disables that configuration flag (AutoJoinSteamGroup), regular players, one-time idlers, and community members remain on the roster alongside automated farming fleets. Furthermore, a Steam ban carries no public reason field. The finding is an empirical rate across this software-associated population, not an individual verdict on any single account. Aggregate, reproducible: asf-ban-census-2026-09-23.json.
Only 8.5% of the banned accounts declare a country at all. Among the ones that do, the largest single share is the United States.
A Steam profile carries an optional country field. 100,413 of the 1,182,811 banned accounts fill it in — 8.5%.† Among that minority the United States leads at 15.5%, then Ukraine, China and Russia, and the rest runs out into a long tail across Europe, Latin America and South-East Asia. Nobody verifies the field: Afghanistan sits in the top ten, which is roughly what it is worth. The other 91.5% are omitted rather than assigned, so what follows is the shape of a self-selected minority — not a map of the fleet, and not a basis for saying where it is run from.
Top 15 countries among the 100,413 banned accounts that expose a public location. Share is of that located subset. Hover for counts.
Location read from GetPlayerSummaries for the banned subset. The field is set by the account holder and verified by nobody, so a VPN exit or a joke entry counts the same as a real one.
Dated by account-creation time, the banned population is recent, not a leftover from the trading-card era.
If this were an old mess being slowly cleaned up, the banned accounts would skew to the early trading-card years. They do the opposite. Of the 1,132,852 banned accounts that expose a creation date, 70% were created in 2020 or later, and nearly half since 2023. The count rises from 59,175 created in 2017 to a peak of 227,030 created in 2024 alone. Every one of them cleared the $5.00 store gate before it could receive a card, so the entry fee was collected on each.
Account-creation year of the 1,132,852 banned members that expose a creation date. The mass is 2020–2026, not the 2013–2015 trading-card era. Peak: 2024. Hover for counts.
Creation dates from GetPlayerSummaries for the banned subset, deduplicated by SteamID64. A creation date is when the account was made, not when it was banned or when it began farming; it is read here as evidence of how recent the population is, not of intent. Three things the chart does not show:
The obvious defence for 88.97% untouched is a backlog. The dated bans say otherwise.
Account age at the moment of the ban, for the 293,750 banned accounts that carry both a creation date and a dated last ban. Over half were caught inside the first year.
This is the answer to the obvious defence. If 88.97% of the fleet is untouched because enforcement is simply slow — a backlog working its way through — then the bans that do land would skew old. They do the opposite: the median banned account was nine and a half months old, and a third — 33.7% — were caught inside six months. When Valve acts on this population, it acts fast. The nine in ten still running are not waiting in a queue.
Age at ban = creation date from GetPlayerSummaries subtracted from the ban date implied by DaysSinceLastBan, read on 2026-09-23. It is computed only where both values are present: 293,750 of the 1,200,011 banned accounts. The remaining accounts report no days-since value and are omitted, not assumed — so read this as the shape of the dated subset, not of every ban.
Of the 11% that carry any ban, look at which ban.
A ban is not a ban. Of the 1,200,011 accounts Valve has flagged, 99.8% carry a community ban — a restriction on forums, profiles and groups that does nothing to a trading bot. The account can still receive cards, still trade, still move inventory. 22.2% carry at least one game ban and 3.9% a VAC ban; the categories overlap, so they do not sum. The one sanction that actually severs trading, the economy / trade ban, sits on 5,214 accounts. We are not resting the case on that number — trade bans are rare across Steam generally. The point of this chart is simpler: even the fraction Valve does touch keeps trading.
Each ban type as a share of all 10,880,610 scanned members. Accounts can carry more than one. Hover for counts.
A Steam ban carries no reason field, so none of these bans is attributed to card farming specifically — this is the mix of ban types present, set beside what each type actually restricts. Community ban does not stop trading; the economy / trade ban does. Aggregate: asf-ban-census-2026-09-23.json.
A second, independent list — community-confirmed fraud — measured against the same official API.
The automation population could be argued away as harmless idlers. The second population cannot. SteamRep is a long-running community reputation database; an account marked BANNED there has been adjudicated a scammer by that community, with documented victim proof and transaction chat logs. We completed a 100% census of all 53,558 profiles carrying that BANNED status against Steam's official Web API (ISteamUser/GetPlayerBans/v1) — and counted only the sanctions that address scamming. A VAC ban is for software cheating in a game (aimbots, wallhacks); it does nothing to a trader and is not evidence Valve acted on fraud, so it is not counted as fraud enforcement.
Across the entire 53,558-account census, 68.63% — 36,757 adjudicated scammers — carry ZERO scam-relevant bans from Valve: they can still trade, still send friend requests to new victims, still message. Only 31.37% were met with either a community or trade ban, and the trade ban — the one action that actually severs inventory transfers — sits on just 18.27% (9,783 accounts). An additional 12.86% (6,888 accounts) carry VAC or game bans purely for in-game cheating, while remaining completely unrestricted in Steam trading.
All 53,558 SteamRep-BANNED profiles, split by whether Valve applied a community or trade ban. VAC and game bans are excluded — they do not address fraud. Hover for counts.
SteamRep is a third-party, community-adjudicated reputation database, not a Valve determination — which is the point: external evidence, measured against what Valve itself did. All 53,558 BANNED profiles from the canonical export were cross-referenced directly against ISteamUser/GetPlayerBans/v1. "Scam-relevant" means a community or economy/trade ban; VAC (12.86%) and game bans (3.86%) are counted separately because they address cheating, not fraud. Complete audit dataset: steamrep_steam_bans.csv.
These are not the same people and not the same conduct. One is ten million automation accounts; the other is fifty-three thousand adjudicated scammers. What they share is the shape of the result: nine in ten farm accounts are never touched, and over two-thirds (68.63%) of confirmed scammers meet no sanction that addresses fraud. The community ban Valve does hand out freely restricts forums and profiles, not trading. The report button, the reputation database, the ban that stops a scam: all present, all under-used. Set that beside the archived refusals in the main dossier, where the same company permanently removes a single account and cites commercial use and automation as the reason — the clause is not dormant, it is pointed.
Statistical sampling logic and mathematical proof: why taking any other 100,000 accounts yields the exact same result.
Valve's primary defense to regulators and consumer protection bodies has been that automated scams and farming are carried out by ephemeral, free "throwaway" accounts that are quickly banned. Under Valve's published rules, an account is Limited until it spends at least $5.00 USD in the Steam store, which locks it out of trading, friend requests, and API features. We audited this claim directly by testing the monetization status of a representative, mathematically rigorous sample of 100,000 banned Steam accounts drawn from our 1,182,811 profile metadata corpus.
Drawn from N = 1,182,811 unique banned Steam accounts using Simple Random Sampling Without Replacement via a cryptographically uniform PRNG. Every account had an identical inclusion probability of P = n / N ≈ 8.45%, completely eliminating survivorship and activity bias.
Applying the Finite Population Correction (FPC) formula at a 99% Confidence Interval (Z = 2.576):MoE = Z × √[p(1-p)/n] × √[(N-n)/(N-1)] ≈ ±0.39%
Because the margin of error is bounded under ±0.4% at 99% confidence, drawing any other independent 100,000-account sample is mathematically guaranteed to fall within 63.06% ± 0.39% (62.67% to 63.45%). The sample mirrors the entire 1.18 million banned population.
The audit results completely demolish the "throwaway bot" defense. Of the 100,000 accounts audited via Steam's legacy XML profile interface: 61,490 accounts (63.06% of classified accounts) are Non-Limited (Paid) accounts having spent at least $5.00 on Steam. Only 36,021 (36.94%) are limited free accounts, while 2,489 were unconfigured or private.
Empirical audit of 100,000 randomly sampled banned accounts. Over 63% have cleared Valve's paid financial gate. Hover for counts.
Why would 63% of the sampled banned accounts be paid accounts that spent real money in Valve's store? Botnet operators do not spend millions out of pocket on disposable accounts. This ratio highlights the volume of compromised and hijacked accounts circulating through the ecosystem.
This scale is directly observable on underground platforms: on marketplaces like LZT Market (Lolzteam), over 550,000 Steam accounts (stealer logs, phishing captures, credential stuffing) are actively listed for sale at any given moment (phishdestroy.io/live-intel). The market authenticates buyers and sellers using Valve's own Steam OpenID, and neither lzt.market nor lolz.team is blocked by Valve's Link Filter.
As established in Section 06, Valve does not proactively eliminate this fraudulent activity: 68.63% of community-confirmed scammers carry zero scam bans on Steam, and 88.97% of the automation fleet remains completely untouched. The accounts circulate freely, authenticate through official APIs, and trade across the ecosystem with near-total impunity.
Sampling artifact: banned_100k_slice.csv (100,000 accounts). Audit outputs: slice_100k_limited_results.csv. Full analytical report: BANNED_100K_FORENSIC_ANALYSIS.md.
Three public populations, one public API, rigorous statistical sampling.
| Population | Where it came from | What was read |
|---|---|---|
| ASF members 10,880,610 | The public member list of the ArchiSteamFarm Steam Community group. Steam's group page shows 5,360,499 members (non-limited only); the API member list returned 10,880,610 total entries including limited accounts. The group is auto-joined on first run unless the user explicitly disables the default group-join setting in configuration (AutoJoinSteamGroup / Statistics), meaning the member list includes commercial bot clusters, individual card idlers, and regular gamers who left the default option enabled. | GetPlayerBans for every member (ban flags); GetPlayerSummaries for creation dates and country. Batched at 100 IDs, deduplicated by SteamID64. |
| 100k Banned Accounts SRSWOR Slice 100,000 | A representative sample drawn uniformly at random without replacement from the 1,182,811 banned accounts corpus ($P = 8.45\%$). Finite Population Correction yields a margin of error of $\pm 0.39\%$ at a 99% confidence level ($Z = 2.576$). | Monetization and limited account status (isLimitedAccount) audited via Steam's legacy XML profile interface (?xml=1) over distributed proxy pipelines. Confirms 63.06% paid status. |
| SteamRep scammers 53,558 | Complete 100% census of all profiles marked BANNED in the SteamRep community reputation database (53,558 total profiles). Zero sampling or filtering applied. | ISteamUser/GetPlayerBans/v1 for all 53,558 accounts, cross-tabulated against SteamRep status. Confirms 68.63% walk free of any scam bans. |
Only empirical rates and statistically bounded estimates. In the automation population, 88.97% carry no ban; the banned subset skews recent (70% since 2020). In the 100,000-account empirical audit, 63.06% of banned accounts are paid (non-limited), with a mathematically proven margin of error under ±0.4%. Among 53,558 confirmed scammers, 68.63% carry no scam-relevant ban. Every figure is reproducible from Valve's own Web API.
A Steam ban carries no reason field, so no ban here is attributed to a specific cause. Group membership is not proof that an account is an automated bot: because ASF enables group-joining by default unless manually unticked, the population mixes commercial multi-account farms with personal accounts of ordinary players and hobbyists. Country is self-reported, and a SteamRep listing is a community adjudication, not a court finding. No individual is named.
Valve holds the same API, its own database behind it, and the ban reasons this scan cannot see. It can run this query across the whole ID space in seconds. The measurement that lets an outside party check an enforcement claim against a number has never been published by the one party that can produce it at full fidelity. This is the outside estimate, offered so the question can be put with a figure attached.
Read next: the card economy this population runs → · support access and theft → · the full dossier →