fbpx

Reduce risks of user sign-up fraud and SMS pumping with Amazon Cognito user pools

If you have a customer facing application, you might want to enable self-service sign-up, which allows potential customers on the internet to create an account and gain access to your applications. While it’s necessary to allow valid users to sign up to your application, self-service options can open the door to unintended use or sign-ups. Bad actors might leverage the user sign-up process for unintended purposes, launching large-scale distributed denial of service (DDoS) attacks to disrupt access for legitimate users or committing a form of telecommunications fraud known as SMS pumping. SMS pumping is when bad actors purchase a block of high-rate phone numbers from a telecom provider and then coerces unsuspecting services into sending SMS messages to those numbers.

Amazon Cognito is a managed OpenID Connect (OIDC) identity provider (IdP) that you can use to add self-service sign-up, sign-in, and control access features to your web and mobile applications. AWS customers who use Cognito might encounter SMS pumping if SMS functions are enabled to send SMS messages, for example, perform user phone number verification during the registration process, to facilitate SMS multi-factor authentication (MFA) flows, or to support account recovery using SMS. In this blog post, we explore how SMS pumping may be perpetrated and options to reduce risks, including blocking unexpected user registration, detecting anomalies, and responding to risk events with your Cognito user pool.

Cognito user sign-up process

After a user has signed up in your application with an Amazon Cognito user pool, their account is placed in the Registered (unconfirmed) state in your user pool and the user won’t be able to sign in yet. You can use the Cognito-assisted verification and confirmation process to verify user-provided attributes (such as email or phone number) and then confirm the user’s status. This verified attribute is also used for MFA and account recovery purposes. If you choose to verify the user’s phone number, Cognito sends SMS messages with a one-time password (OTP). After a user has provided the correct OTP, their email or phone number is marked as verified and the user can sign in to your application.

Figure 1: Amazon Cognito sign-up process

Figure 1: Amazon Cognito sign-up process

If the sign-up process isn’t protected, bad actors can create scripts or deploy bots to sign up a large number of accounts, resulting in a significant volume of SMS messages sent in a short period of time. We dive deep into prevention, detection, and remediation mechanisms and strategies that you can apply to help protect against SMS pumping based on your use case.

Protect the sign-up flow

In this section, we review several prevention strategies to help protect against SMS sign-up frauds and help reduce the amount of SMS messages sent to bad actors.

Implement bot mitigation

Implementing bot mitigation techniques, such as CAPTCHA, can be very effective in preventing simple bots from pumping user creation flows. You can integrate a CAPTCHA framework on your application’s frontend and validate that the client initiating the sign-up request is operated by a human user. If the user has passed the verification, you then pass the CAPTCHA user response token in ClientMetadata together with user attributes to an Amazon Cognito SignUp API call. As part of the sign-up process, Cognito invokes an AWS Lambda function called pre sign-up Lambda trigger, which you can use to reject sign-up requests if there isn’t a valid CAPTCHA token presented. This will slow down bots and help reduce unintended account creation in your Cognito user pool.

Validate phone number before user sign-up

Another layer of mitigation is to identify the actor’s phone number early in your application’s sign-up process. You can validate the user provided phone number in the backend to catch incorrectly formatted phone numbers and add logic to help filter out unwanted phone numbers prior to sending text messages. Amazon Pinpoint offers a Phone Number Validate feature that can help you determine if a user-provided phone number is valid, determine phone number type (such as mobile, landline, or VoIP), and identify the country and service provider the phone number is associated with. The returned phone number metadata can be used to decide whether the user will continue the sign-up process and send an SMS message to that user. Note that there’s an additional charge for using the phone number validation service. For more information, see Amazon Pinpoint pricing.

To build this validation check into the Amazon Cognito sign-up process, you can customize the pre sign-up Lambda trigger, which Cognito uses to invoke your code before allowing users to sign-up and sending out an SMS OTP. The Lambda trigger invokes the Amazon Pinpoint phone number validate API, and based on the validation response, you can build a custom pattern that fits your application to continue or reject the user sign-up. For example, you can reject user sign-ups with VoIP numbers or reject users who provide a phone number that’s associated with countries that you don’t operate in, or even reject certain cellular service providers. After you reject a user sign-up using the Lambda trigger, Cognito will deny the user sign-up request and will not invoke user confirmation flow nor send out an SMS message.

Example validation command using AWS CLI

aws pinpoint phone-number-validate --number-validate-request PhoneNumber=+155501001234

When you send a request to the Amazon Pinpoint phone number validation service, it returns the following metadata about the phone number. The following example represents a valid mobile phone number data set:

{
    "NumberValidateResponse": {
        "Carrier": "ExampleCorp Mobile",
        "City": "Seattle",
        "CleansedPhoneNumberE164": "+155501001234",
        "CleansedPhoneNumberNational": "55501001234",
        "Country": "United States",
        "CountryCodeIso2": "US",
        "CountryCodeNumeric": "1",
        "OriginalPhoneNumber": "+155501001234",
        "PhoneType": "MOBILE",
        "PhoneTypeCode": 0,
        "Timezone": "America/Seattle",
        "ZipCode": "98109"
    }
}

Note that PhoneType includes type MOBILE, LANDLINE, VOIP, INVALID, or OTHER. INVALID phone numbers don’t include information about the carrier or location associated with the phone number and are unlikely to belong to actual recipients. This helps you decide when to reject user sign-ups and reduces SMS messages to undesired phone numbers. You can see details about other responses in the Amazon Pinpoint developer guide.

Example pre sign-up Lambda function to block user sign-up except with a valid MOBILE number

The following pre sign-up Lambda function example invokes the Amazon Pinpoint phone number validation service and rejects user sign-ups unless the validation service returns a valid mobile phone number.

import { PinpointClient, PhoneNumberValidateCommand } from "@aws-sdk/client-pinpoint"; // ES Modules import

const validatePhoneNumber = async (phoneNumber) => {
  const pinpoint = new PinpointClient();
  const input = { // PhoneNumberValidateRequest
    NumberValidateRequest: { // NumberValidateRequest
      PhoneNumber: phoneNumber,
    },
  };
  const command = new PhoneNumberValidateCommand(input);
  const response = await pinpoint.send(command);

  return response;
};

const handler = async (event, context, callback) => {

  const phoneNumber = event.request.userAttributes.phone_number;
  const validationResponse = await validatePhoneNumber(phoneNumber);

  if (validationResponse.NumberValidateResponse.PhoneType != "MOBILE") {
    var error = new Error("Cannot register users without a mobile number");
    // Return error to Amazon Cognito
    callback(error, event);
  }
  // Return to Amazon Cognito
  callback(null, event);
};

export { handler };

Use a custom user-initiated confirmation flow or alternative OTP delivery method

In your user pool configurations, you can opt out of using Amazon Cognito-assisted verification and confirmation to send SMS messages to confirm users. Instead, you can build a custom reverse OTP flow to ask your users to initiate the user confirmation process. For example, instead of automatically sending SMS messages to a user when they sign up, your application can display an OTP and direct the user to initiate the SMS conversation by texting the OTP to your service number. After your application has received the SMS message and confirmed the correct OTP is provided, invoke a service such as a Lambda function to call the AdminConfirmSignUp administrative API operation to confirm user, then call AdminUpdateUserAttributes to set the phone_number_verified attribute as true to indicate that the user phone number is verified.

You can also choose to deliver an OTP using other methods, such as email, especially if your application doesn’t require the user’s phone number. During the user sign-up process, you can configure a custom SMS sender Lambda trigger in Amazon Cognito to send a user verification code through email or another method. Additionally, you can use the Cognito email MFA feature to send MFA codes through email.

Detect SMS pumping

When you’re considering the various prevention options, it’s important to set up detection mechanisms to identify SMS pumping as they arise. In this section, we show you how to use AWS CloudTrail and Amazon CloudWatch to monitor your Amazon Cognito user pool and detect anomalies that could lead to SMS pumping. Note that building detection mechanism based on anomalies requires knowing your average or baseline traffic and the difference in metrics that represent regular activity and metrics that can indicate unauthorized or unintended activity.

Service quotas dashboard and CloudWatch alarms

Bad actors may attempt to leverage either the sign-up confirmation or the reset password functionality of Amazon Cognito. As shown previously in Figure 1, when a new user signs up to your Cognito user pool, the SignUp API operation is invoked. When the user provides the OTP confirmation code, the ConfirmSignUp API operation is invoked. The call rate of both APIs is tracked collectively under Rate of UserCreation requests under Amazon Cognito service in the service quotas dashboard.

You can set up Amazon CloudWatch alarms to monitor and issue notifications when you’re close to a quota value threshold. These alarms could be an early indication of a sudden usage increase, and you can use them to triage potential incidents.

Additionally, when your services are sending SMS messages, those transactions count towards the Amazon Simple Notification Service (Amazon SNS) service quota. You should set up alarms to monitor the Transactional SMS Message Delivery Rate per Second quota and the SMS Message Spending in USD quota.

CloudTrail event history

When bad actors plan SMS pumping, they are likely attempting to trick you to send as many SMS messages as possible rather than completing the user confirmation process. Under the context of a user sign-up event, you might notice in the CloudTrail event history that there are more SignUp and ResendConfirmationCode events—which send out SMS messages—than ConfirmSignUp operations; indicating a user has initiated but not completed the sign-up process. You can use Amazon Athena or CloudWatch Logs Insights to search and analyze your Amazon Cognito CloudTrail events and identify if there’s a significant reduction in finishing the user sign-up process.

Figure 2: SignUp API logged in CloudTrail event history

Figure 2: SignUp API logged in CloudTrail event history

Similarly, you can apply this observability towards the user password reset flow by analyzing the ForgotPassword API and ConfirmForgotPassword API operations for deviations.

Note that the slight deviations in user completion flow in the CloudTrail event history alone might not be an indication of unauthorized activity, however a substantial deviation above the regular baseline might be a signal of unintended use.

Monitor excessive billing

Another opportunity for detecting and identifying unauthorized Amazon Cognito activity is by using AWS Cost Explorer. You can use this interface to visualize, understand, and manage your AWS costs and usage over time, which might assist by highlighting the source of excessive billing in your AWS account. Be aware that charges in your account can take up to 24 hours to be displayed, so while this method can help provide some assistance in identifying SMS pumping activity, it should only be used as a supplement to other detection methods.

To use Cost Explorer:

  1. Open the AWS Management Console, and go to Billing and Cost Management.
  2. In the navigation pane, under Cost Analysis, choose Cost Explorer.
  3. In the Cost and Usage Report, under Report Parameters, select Date Range to include the start and end date of the time period that you want to apply a filter to. In Figure 3 that follows, we use an example date range between 2024-07-03 and 2024-07-17.
  4. In the same Report Parameter area, under Filters, for Service, select SNS (Simple Notification Service). Because Amazon Cognito uses Amazon SNS for delivery of SMS messages, filtering on SNS can help you identify excessive billing.

Figure 3: Reviewing billing charges by service

Figure 3: Reviewing billing charges by service

Apply AWS WAF rules as mitigation approaches

It’s recommended that you apply AWS WAF with your Amazon Cognito user pool to protect against common threats. In this section, we show you a few advanced options using AWS WAF rules to block or throttle specific bad actor’s traffic when you have observed irregular sign-up attempts and suspect they were part of fraudulent activities.

Target a specific bad actor’s IP address

When building AWS WAF remediation strategies, you can start by building an IP deny list to block traffic from known malicious IP addresses. This method is straightforward and can be highly effective in preventing unwanted access. For detailed instructions on how to set up an IP deny list, see Creating an IP set.

Target a specific phone number area code regex pattern

In an SMS pumping scheme, bad actors often purchase blocks of cell phone numbers from a wireless service provider and use phone numbers with the same area code. If you observe a pattern and identify that these attempts use the same area code, you can apply an AWS WAF rule to block that specific traffic.

To configure an AWS WAF web ACL to block using an area code regex pattern:

  1. Open the AWS WAF console.
  2. In the navigation pane, under AWS WAF, choose WAF ACLs.
  3. Choose Create web ACL. Under Web ACL details, select Regional resources, and select the AWS Region as your Amazon Cognito user pool. Under Associated AWS resources, select Add AWS resources, and choose your Cognito user pool. Choose Next.
  4. On the Add rules and rule groups page, choose Add rules, Add my own rules and rule groups, and Rule builder.
  5. Create a rule in Rule builder.
    1. For If a request, select matches the statement.
    2. For Inspect, select Body.
    3. For Match type, select Matches regular expression.
    4. For Regular expression, enter a match for the observed pattern. For example, the regular expression ^303|^+1303|^001303 will match requests that include the digits 303, +1303, or 001303 at the beginning of any string in the body of a request:

    Figure 4: Creating a web ACL

    Figure 4: Creating a web ACL

  6. Under Action, choose Block. Then, choose Add rule.
  7. Continue with Set rule priority and Configure metrics, then choose Create web ACL.

Be aware that this method will block all user sign-up requests that contain phone numbers matching the regex pattern for the target area code and could prevent legitimate users whose numbers match the defined pattern from signing up. For example, the rule above will apply to all users with phone numbers starting with 303, +1303, or 001303. You should consider implementing this method as an as-needed solution to address an ongoing SMS pumping attack.

Target a specific bad actor’s client fingerprint

Another method is to examine an actor’s TLS traffic. If your application UI is hosted using Amazon CloudFront or Application Load Balancer (ALB), you can build AWS WAF rules to match the client’s JA3 fingerprint. The JA3 fingerprint is a 32-character MD5 hash derived from the TLS three-way handshake when the client sends a ClientHello packet to the server. It serves as a unique identifier for the client’s TLS configuration because various attributes such as TLS version, cipher suites, and extensions are derived to calculate the fingerprint, allowing for the unique detection of clients even when the source IP and other commonly used identification information might have changed.

Fraudulent activities, such as SMS pumping, are typically carried out using automated tools and scripts. These tools often have a consistent SSL/TLS handshake pattern, resulting in a unique JA3 fingerprint. By configuring an AWS WAF web ACL rule to match the JA3 fingerprint associated with this traffic, you can identify clients with a high degree of accuracy, even if they change other attributes, such as IP addresses.

AWS WAF has introduced support for JA3 fingerprint matching, which you can use to identify and differentiate clients based on the way they initiate TLS connections, enabling you to inspect incoming requests for their JA3 fingerprints. You can build the remediation strategy by first evaluating AWS WAF logs to extract JA3 fingerprints for potential malicious hosts, then proceed with creating rules to block requests where the fingerprint matches the malicious JA3 fingerprint associated with previous attacks.

To configure an AWS WAF web ACL to block using JA3 fingerprint matching for CloudFront resources:

  1. Open the AWS WAF console.
  2. In the navigation pane, under AWS WAF, choose WAF ACLs.
  3. Choose Create web ACL. Under Web ACL details, select Amazon CloudFront distributions. Under Associated AWS resources, select Add AWS resources, and select your CloudFront distribution. Choose Next.
  4. On the Add rules and rule groups page, choose Add rules, Add my own rules and rule groups, and Rule builder.
  5. In Rule builder:
    1. For If a request, select matches the statement.
    2. For Inspect, select JA3 fingerprint.
    3. For Match type, keep Exactly matches string.
    4. For String to match, enter the JA3 fingerprint that you want to block.
    5. For Text transformation, choose None.
    6. For Fallback for missing JA3 fingerprint, select a fallback match status for cases where no JA3 fingerprint is detected. We recommend choosing No match to prevent unintended traffic blocking.
    7. If you need to block multiple JA3 fingerprints, include each one in the rule and for If a request select matches at least one of the statements (OR).

      Figure 5: Creating an AWS WAF statement for a JA3 fingerprint

      Figure 5: Creating an AWS WAF statement for a JA3 fingerprint

    8. Under Action, select Block, and choose Add rule. You can choose other actions such as COUNT or CAPTCHA that suit your use case.
  6. Continue with Set rule priority and Configure metrics, then choose Create web ACL.

Note that JA3 fingerprints can change over time due to the randomization of TLS ClientHello messages by modern browsers. It’s important to dynamically update your web ACL rules or manually review logs to update the JA3 fingerprint search string in your match rule when applicable.

AWS WAF remediation considerations

These AWS WAF remediation approaches help to block potential threats by providing mechanisms to filter out malicious traffic. It’s essential to continually review the effectiveness of these rules to minimize the risk of blocking legitimate sources and make dynamic adjustments to the rules when you detect new bad actors and patterns.

Summary

In this blog post, we introduced mechanisms that you can use to detect and protect your Amazon Cognito user pool against unintended user sign-up and SMS pumping. By implementing these strategies, you can enhance the security of your web and mobile applications and help to safeguard your services from potential abuse and financial loss. We suggest that you apply a combination of these prevention, detection, and mitigation approaches to protect your Cognito user pools.

 
If you have feedback about this post, submit comments in the Comments section below. If you have questions about this post, contact AWS Support.

Edward Sun
Edward Sun

Edward is a Security Specialist Solutions Architect focused on identity and access management. He loves helping customers throughout their cloud transformation journey with architecture design, security best practices, migration, and cost optimizations. Outside of work, Edward enjoys hiking, golfing, and cheering for his alma mater, the Georgia Bulldogs.
Steve de Vera
Steve de Vera

Steve is a manager in the AWS Customer Incident Response Team (CIRT). He is passionate about American-style BBQ and is a certified competition BBQ judge. He has a dog named Brisket.
Tony Suarez
Tony Suarez

Tony Suarez is a San Diego, CA based Solutions Architect with over 15 years of experience in IT operations. As a member of the AWS VMware technical field community, Tony enjoys helping customers solve challenging problems in innovative ways. Enabling customers to efficiently manage, automate, and orchestrate large-scale hybrid infrastructure projects is Tony’s passion.