How Stripe AI Fraud Detection Works

11 Oct 2026

Stripe's Radar scores every payment with machine learning and gives it a risk level. Here is what the score means and what you can change.

A customer checks out. Before the payment settles, Stripe's AI has already looked at it, scored it, and decided how risky it is. Here is how Stripe AI fraud detection works, in plain terms, and what it does and does not do for your store.

What Stripe's AI fraud detection is

The tool is called Radar. Stripe's guide calls it a suite of fraud-fighting tools built into Stripe that needs no additional integration work. There is no additional integration work.

Radar pairs AI algorithms with a rules engine that you can edit. It assesses the risk of each payment. The AI half is the part this article is about. For the wider picture, see ai fraud detection.

How the AI works

Radar uses machine learning. Stripe's guide calls machine learning a subfield of artificial intelligence. The models learn from data.

The models learn from dispute data, customer information, and browsing data. This helps them spot risky transactions and flag fewer good ones. They evaluate hundreds of risk factors when scoring a charge. They also use signals from across the Stripe network as a reference for spotting patterns.

Radar checks transactions, accounts, and customers in real time. Its AI algorithms assess the risk of fraud. And the models do not sit still. They can be retrained on new data so they stay up to date and better detect emerging fraud patterns. That is why machine learning fits real-time fraud detection. It can spot patterns and odd behavior that point to fraud.

Risk scores and risk levels

Every payment gets a numerical risk score between 0 and 99, where 0 is the lowest risk and 99 is the highest. A score of 65 or above indicates elevated risk. A score of 75 or above indicates high risk.

Each score also gets a risk level: normal, elevated, highest, or not assessed. How scores work in general is covered in fraud scoring.

Stripe can also explain itself. It also gives agentic risk insights. These explain the risk level, such as odd transaction patterns or behavior that matches past fraud. So when a payment is flagged, you can see what caught the model's eye.

What happens automatically

When Radar believes a payment is likely to be fraudulent, it labels it high risk and blocks it without you doing anything. Payments at the elevated level go through unless you step in. Normal risk payments go through.

So yes, Stripe blocks risky payments on its own. You do not need a rule for that.

Customizing the AI's decisions

The out-of-the-box settings are only where you begin. You can write custom rules for more control over which payments to review, allow, or block. You can set the criteria for connected account and transaction rules. That way they fit the risks of your business.

Risk settings let you balance authorization and fraud on your account by using risk controls. Risk controls use AI to protect your account. They block payments that might lead to fraudulent disputes or early fraud warnings. Three settings exist:

Maximize protection puts safety first. It stops payments that seem headed for early fraud warnings.
Balance risk and revenue splits the difference, keeping some fraud protection while accepting some risk.
Maximize revenue puts sales first. It prioritizes revenue by blocking high-risk payments.

You can also teach the AI. Refund and report payments you believe are fraudulent. That feedback sharpens the algorithms and risk evaluations over time. When you mark a payment as fraudulent, its email address and card fingerprint land on your block lists.

Bot detection and the bot score

Radar also evaluates whether a payment was likely made by a bot. Radar Pro includes the bot risk control and the bot score.

The bot score ranges between 0 and 99, where 99 is the highest likelihood a bot made the payment. A high bot score indicates a bot likely made the payment, not that the payment is fraudulent. The two scores answer different questions, so read them separately.

Radar plans at a glance

Radar offers several plans, based on a business's needs for real-time fraud protection. Check Stripe's current pricing page for costs.

Radar Lite offers AI-based fraud prevention for payments and card testing, plus fraud alerts.
Radar Standard offers out-of-the-box fraud protection for all payment methods. It detects and prevents transaction fraud, and spots fraudulent accounts.
Radar Plus lets you write and backtest custom rules, set risk tolerance, and get advanced analytics.
Radar Pro guards against new fraud threats and customer abuse.

For how a team works with Radar's review queue and rules, see stripe radar for fraud teams.

What Stripe's AI doesn't do

Radar does not catch everything, and Stripe says so plainly. Payments with normal risk can still turn out to be fraudulent. A green score is a signal, not a promise.

Stripe is also clear that the payments you accept are your call, and your responsibility. That covers payments that get disputed later, or turn out to be fraud. A fraudulent sale can come back as a chargeback, and fraud and disputes go hand in hand. The more you do to reduce fraud, the lower your dispute rate.

That is why the score is a signal to act on, not a verdict to hide behind. Review what Radar flags, report what you know was fraud, and write rules that fit your store. Radar does the first pass on card not present fraud detection. The judgment calls stay with you.