Back to Blog

How to stop free CI minutes abuse

Free CI minutes abuse fills shared runners and makes paying teams wait. The queue arithmetic, what each basic fix misses, and what to do first.

Why is every paying team's build stuck on "Waiting for a runner"? Because 300 of your 400 shared runners are busy with jobs from 60 free accounts, and all 60 were created last night.

That's a scenario, so swap in your own numbers as you read. Say you run a hosted CI service. Your free plan gives each account 2,000 minutes a month and five concurrent jobs, and a runner-minute costs you $0.005 all in (an assumption covering compute, network and logs). Each of those 60 accounts starts five jobs that run up to your six-hour job limit, then more until its minutes are gone. This is free CI minutes abuse, and it stays under every limit you set.

60 accounts x 5 concurrent jobs   = 300 runners held
2,000 minutes / 5 concurrent jobs = 400 minutes, or 6 hours 40 minutes
60 accounts x 2,000 minutes       = 120,000 runner-minutes
120,000 runner-minutes x $0.005   = $600 for one batch

The batch started at 6 a.m., so it holds those runners until 12:40 p.m. Your paying teams start merging around 9:30. At their peak they need 180 runners and their average job takes 10 minutes, so 18 new jobs arrive every minute. They've got 100 runners left, which clear 10 jobs a minute. The line grows by 8 jobs a minute, and it's the same line for everybody.

Wait for a new paying job, illustrative0204060801009:3010:0010:3011:0011:30Minutes in lineTime of day
Wait for a new paying job, illustrative0501009:3010:0010:3011:0011:30Minutes in lineTime of day

A customer pushes a hotfix at 10:30, behind 480 other jobs (60 minutes × 8), so it can't start until 11:18. At 11:05 their engineer writes in: "Our hotfix has said Waiting for a runner for 35 minutes. Is CI down?" Your status page is green, and it's right. Nothing is down. Twelve blocked engineers for 48 minutes is 9.6 hours of their time (12 × 48 ÷ 60), and the fix still isn't live.

Now the money. One batch every weekday is 20 batches a month:

  • 20 batches × $600 = $12,000 a month, spent on 1,200 accounts that will never pay
  • Against 300 paying teams at $149, or $44,700 a month, that's about 27% of revenue
  • A month's credit for the 25 teams who complained is another $3,725 (25 × $149)
  • With $600,000 in the bank and a $40,000 monthly burn you had 15 months. At $52,000 you have about 11 and a half

The costs also outlast the signup. All 1,200 accounts get their 2,000 minutes back on the 1st, so another $12,000 is available without one new signup (1,200 × 2,000 × $0.005). If each left 8 GB of caches and artifacts, that's 9,600 GB, and 90 days of it at an assumed $0.03 per GB per month costs $864.

So now you choose. Buying 300 more always-on runners costs 300 × 43,800 minutes a month × $0.005 = $65,700 a month, more than all your paying teams pay you. Or you cut the free plan, and tell the maintainer who's run their test suite on it for three years that it's gone because of accounts they've never heard of. That free plan is how most of your 300 teams found you.

Why free CI minutes attract this

A free plan for a note-taking app gives away note-taking. A free CI plan gives away a Linux machine by the hour, and a job can run almost anything you can put in a config file. Some people want the machine and don't care about the build. GitLab said as much in May 2021, when it began asking new free users for a valid card before they could use shared runners, because of cryptomining on free pipeline minutes.

Plenty of it is duller than that. A five-person agency puts each client's repository on its own free account so nobody ever needs a second plan. Each account looks like a small, real project, because it is one.

What makes CI different from most free plans is the queue. Minutes are a budget, but runners are a fixed pool, so a free job costs you money and also takes a slot a paying job needed.

5 jobs each, 300 in all300 runners takenMerge to main, 20 jobsWaiting for a runnerMinutes gone after 6h 40mLine grows by 8 a minute60 free accountsShared queue400 runnersPaying team60 free accountsShared queue400 runnersPaying team
5 jobseach,300 in all300runnerstakenMerge to main, 20jobsWaiting for a runnerMinutes gone after 6h40mLine grows by 8 a minute60freeaccountsSharedqueue400runnersPayingteam60freeaccountsSharedqueue400runnersPayingteam

The obvious fix is a limit per account, and you already have two of them. Those 60 accounts are under both.

Fix the queue first

All of this is worth shipping this week.

Split the pool. Cap free jobs at 40 runners and keep 360 for paying teams, twice their 180 peak, so the line in that chart never forms. It also puts a ceiling on the free plan's bill: 40 × 1,440 minutes × $0.005 = $288 a day.

Shorten what free jobs can hold. A one-hour job limit on the free plan frees a runner six times as often. Seven days of artifact retention instead of 90 takes that $864 down to $67.20 (9,600 GB × $0.03 × 7 ÷ 30).

Start new accounts at one concurrent job. Raise it to five after a first payment or 30 days. Sixty accounts now hold 60 runners, not 300. Real test matrices run slower for a while, so tell people why.

Refresh minutes when an account comes back. Grant the monthly allowance on the first job of the month, through one server function that records it. Accounts nobody returns to stop costing you anything.

Your paying teams are safe now, but the free pool isn't. Those 60 accounts still have 120,000 minutes to spend. At one job each they fill all 40 free runners with 20 jobs waiting, and 120,000 ÷ 40 = 3,000 minutes, so the free pool stays full for 50 hours. A developer trying your product for the first time watches their first build sit there and doesn't come back. To do better, fewer of those accounts have to get the minutes at all.

Basic fixes for free CI minutes abuse, and how each one breaks

Most teams go to the signup form next. Gmail ignores dots, so sam.lee@gmail.com and samlee@gmail.com reach the same inbox, and plus-tagged mail like samlee+ci@gmail.com lands there too. That's how email works.

Basic fixWhat it catchesWhat it missesWho it wrongly blocksWhat it costs to keep working
Email verificationTypos and dead addressesAny second real inboxStrict company mail filtersDeliverability work
Disposable-domain blocksKnown throwaway servicesNew domains and ordinary mailboxesMaintainers using privacy forwardingA list someone updates forever
Gmail dots and plus addressingOne Gmail inbox written several waysA second real mailboxPeople who sort their mail with plus addressesUpkeep as mail services change
CAPTCHASimple signup scriptsPeople opening accounts by handScreen reader and privacy tool usersFees and lost signups
Phone requirementCasual second accountsAnyone with another numberPeople without a reliable mobileA charge for every text
Card requirementCasual second accountsAnyone with another cardStudents and hobby maintainersFewer signups, more support
One per IP addressRepeat signups on one connectionAny other networkOffices, campuses and coworking spacesExceptions and tickets
VPN blocksKnown VPN rangesEverything off your listEngineers on company VPNsAnother list to maintain
Rate limitsBursts from one API key60 accounts each under the limitReal teams pushing a big monorepoTuning per endpoint
Usage capsOne account overrunningEach new account's fresh 2,000 minutesAnyone who hits the cap mid-releaseDeciding when builds stop
Manual reviewWhat a person can spotSlow, quiet batchesWhoever looks odd to a tired reviewer1,200 accounts × 1 minute = 20 hours a month

Your own records can tell you whether this account has had its minutes this month. A brand new account has no record at all, so every account in the next batch gets its minutes.

A card requirement falls hardest on the open-source maintainers who make a free tier worth having, and blocking VPNs turns away engineers on company networks, who can bring a paying team with them.

Cut the allowance from 2,000 minutes to 500 and the same 120,000 minutes takes 240 accounts instead of 60, while every honest project gets a quarter of what it had.

Signup checks run once. After that, every job arrives from a push with no form in front of it, so an account that got through keeps its minutes month after month.

Evaluate the claim before the minutes go out

Every row in that table runs out at the same place: an account with no history looks clean. AbuseZero is the abuse prevention API built for that moment. Before your service grants free minutes, your server makes one API call and gets a clear allow or deny. A single check looks at one thing. AbuseZero's multi-layer AI risk analysis examines 140,000+ signals, detects linked claims and flags abuse that single checks miss. We built a proprietary algorithm for this job, with AI that checks the claim again.

Where your domain list goes stale, AbuseZero checks mailbox aliases, disposable domains and mail routing. Where a one-per-IP rule locks out an office, it sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence. Where a CAPTCHA falls short, it blocks spammy bots, browser automation and malicious scripts while allowing genuine AI agents and assistants. It also recognizes repeat devices across new accounts, even when email addresses keep changing. Creating a fresh account isn't enough to erase previous activity.

Your runners, queues and job limits stay in your product, and you don't have to build your own abuse prevention system as well. Eligibility can be once per user or once per calendar month, so a signup allowance and a monthly refresh can each have the right limit. Observe mode shows you what would be denied before anyone is blocked, and the quickstart walks you through your first benefit.

Frequently asked questions

What counts as free CI minutes abuse?

Opening extra accounts to collect free minutes more times than you meant to give them. A real project that uses its whole allowance every month is your free plan doing its job.

Should I remove my free CI tier?

Try the split pool and lower starting concurrency first. Developers tend to find CI tools through free tiers, and removing yours lands on every honest project to reach a few accounts.

How do I stop new accounts claiming free minutes again?

Evaluate each claim when the minutes are granted. AbuseZero's evaluation returns allow or deny under the eligibility rules you set.

Give your paying teams their runners back

Keep reading

Copying your prompt…

Copied the to your clipboard. Give it to your coding agent & it will handle the rest.