Onfido Privacy Policy

Inference & Profiling · Company practice

Facial Similarity Check - Selfie Auto makes the comparison automatically, without human review

Facial Similarity Check - Selfie Auto compares the face displayed on an identity document or reference image with an image of the User's face to verify that they are the same. The notice states that this is done automatically and without any human review.

Supporting quotations

| Categories of Personal Data: ● Image of the User’s face ● Image of the face in the identity document / reference image ● Scans of face geometry | Facial Similarity Check - Selfie Auto Automatically (and without any human review) compares the face displayed on an identity document or reference image with an image of the User’s face, to verify that they are the same. |

Qualifications & exceptions

• The notice expressly says 'without any human review' for this check.

Inference & Profiling · Company practice

Facial Similarity Check - Motion runs automatically with customizable liveness detection and optional audio

Facial Similarity Check - Motion automatically, and without any human review, compares the face displayed on an identity document or reference image with a facial scan captured from a video of the User to verify that they are the same, while also performing customizable liveness detection. Audio processing is optional for this check. Its data includes video of the User with background audio at the Customer's option, the reference face image, transcribed text from the video clip if applicable, and data that may be biometric identifiers or information such as face-geometry scans or voiceprints.

Supporting quotations

| Categories of Personal Data: • Video of the User (including background | Facial Similarity Check - Motion |

audio recording at Customer’s option) ● Image of the face in the identity document / reference image or video ● Transcribed text from the video clip (if applicable) ● Data that may be considered biometric identifiers or biometric information (e.g., scans of face geometry or voiceprints) | Automatically (and without any human review) compares the face displayed on an identity document or reference image with a facial scan captured from a video of the User, to verify that they are the same, while also performing customizable liveness detection. Includes optional processing of audio. |

Qualifications & exceptions

• The check runs automatically and without human review.
• Liveness detection is customizable.
• Background audio recording is at the Customer's option and audio processing is optional.

Inference & Profiling · Company practice

Phone Verification Report checks mobile-number status through mobile network databases and scores fraud risk

The Phone Verification Report searches a range of mobile network operator databases, through a third party, to verify whether a phone number provided by a User is active and reachable. It also considers additional phone intelligence attributes to help the Customer assess the fraud risk associated with the phone number. The listed data is full name, address, mobile number, mobile number network history details, mobile account details and a mobile risk score.

Supporting quotations

| Categories of Personal Data: ● Full name ● Address ● Mobile number ● Mobile number network history details ● Mobile account details ● Mobile risk score | Phone Verification Report Searches a range of mobile network operator databases (via a third party) to verify if a phone number provided by a User is active and reachable; and considers additional phone intelligence attributes to help the Customer to assess the fraud risk associated with the phone number. |

Qualifications & exceptions

• The searches are conducted through a third party.
• The report assists the Customer's fraud-risk assessment; it does not itself state the Customer's decision.

Not stated here

• The notice does not identify the third party or the mobile network databases used.

Inference & Profiling · Company practice

Device Intelligence assesses device, email and phone signals and can infer location and an identity risk score

Device Intelligence helps Customers determine whether a device, email address or phone number has previously been used in relation to suspected fraudulent activity, shows unusual usage patterns, has been manipulated or otherwise indicates the user may not be genuine. At a Customer's request, Entrust and its Data Providers may collect 'passive signals' from the Customer (such as a mobile number or email address) or from a User's device as the User engages with the Customer's website, app or services. The information may include device identifiers, IP address, device information such as operating system and whether the device provides false randomized device or network information or has otherwise been compromised, and how the User interacts with the Services or their device. Examples include upload time, software version, camera name and model, indicators of tampering or emulation, clipboard use and navigation between applications. Entrust and Data Providers may use this information to infer other information such as broad geographical location from the IP address, or to calculate an identity risk score, helping them assess the likelihood of the individual being a genuine User.

Supporting quotations

Device Intelligence. Entrust can help Customers to determine whether a device, email address or phone number has previously been used in relation to suspected fraudulent activity, shows unusual usage patterns, has been manipulated or otherwise indicates that the user may not be genuine. At a Customer’s request, Entrust and our Data Providers may collect ‘passive signals’ from the Customer (for example mobile number or email address) or a User’s device as they engage with the Customer’s website, app or services. Such information may include device identifiers, IP address, information about the device (for example the operating system used, whether the device is providing false randomized device and network information or has otherwise been compromised) and how the User interacts with:

• The Services. For example, information about the upload time, which version of our software was used, the camera name and model used to capture any images and whether there are any indicators that the device has been tampered with or emulated; and • Their device. For example, fraudsters will cut and paste large volumes of information from their clipboard, use this functionality multiple times and otherwise navigate between applications on their device very differently from a genuine User.

Entrust and our Data Providers may also use such information to infer other information about the User, for example their broad geographical location from their IP address, or to calculate an identity risk score.

an individual being a genuine User, assign a risk score and infer certain information such as a User’s broad geographical location from their IP address.

Qualifications & exceptions

• Collection occurs at a Customer's request.
• Passive signals may come from the Customer or from the User's device.
• The notice says Entrust and Data Providers 'may' use the information to infer location or calculate a risk score.
• The listed examples of interaction data are illustrative, not an exhaustive list.

Not stated here

• The notice does not specify the full set of risk-score inputs or how the score is used by a Customer.

Inference & Profiling · Company practice

Device Intelligence - Standard analyzes technical data, device signals, IP address and geolocation

Device Intelligence - Standard analyzes technical data, device signals, IP address, geolocation and other fraud signals to assess the likelihood that the User is genuine. The signals are collected through the SDK or standard tracking technologies such as JavaScript snippets from Users as they interact with the Services. The listed data is broad: a User unique identifier; document number and data from an associated document check; address and IP address type with city/region/country location; check status/outcome and tracking information; Device Intelligence inferences and outputs such as risk score and risk level; device fingerprint and identifiers including audio/video fingerprint, Android/Google Advertising ID and hardware-based identifiers such as MAC address; uploaded images/videos and their metadata/metatags; device hardware and software attributes; behavior during the Services including timing, forms, pointer/touch events, sensor data such as accelerometer and gyroscope; application authenticity such as use of stolen security tokens; and other fraud signals listed in the Onfido API Documentation.

Supporting quotations

| Categories of Personal Data: ● User unique identifier ● Document number and other data extracted from any associated document check ● address and IP address type, including associated city/region/country level location ● Check status/outcome and related tracking information ● Inferences and outputs from Device Intelligence e.g. risk score, risk level; ● Device Fingerprint, device identifiers including audio and video fingerprint, Android/Google Advertising ID (AAID or GAID), hardware-based identifiers (e.g. MAC address) ● Images and videos uploaded by User and any related metadata and metatags ● Information about Users' device including hardware and software attributes (e.g. device type and model, manufacturer, operating system type and version (e.g. iOS or Android), browser type and version, user-agent, navigator, screen details, plugins, fonts, memory, WebGL, battery information, language, time zone, camera name, microphone name, aspect ratio, resolution, frame rate, etc.) ● Information about an end user’s behavior when interacting with the Services, including how the end user uses the Services such as start/stop time, forms/fields completed, pointer and touch events, time zone offset, distraction events, sensor data like accelerometer and gyroscope Application Authenticity - Whether the device is using stolen security tokens ● Other fraud signals listed in the Onfido API Documentation.

Device Intelligence – Standard Analysis of technical data, device signals, IP address, geolocation, and other fraud signals (collected via the SDK or standard tracking technologies (e.g. JavaScript snippets) from Users as they interact with the Services) to assess the likelihood that the User is genuine.

Qualifications & exceptions

• Collection occurs as Users interact with the Services, via the SDK or standard tracking technologies such as JavaScript snippets.
• The listed device attributes are examples ('e.g.') and not an exhaustive list.
• Other fraud signals are defined in the Onfido API Documentation, which is not part of the saved source.

Not stated here

• The Onfido API Documentation listing other fraud signals is not supplied.

Inference & Profiling · Company practice

Mastercard Identity combines submitted data with Mastercard Identity Network signals

Mastercard Identity looks for patterns in the submitted information combined with signals provided by the Mastercard Identity Network to determine if the applicant is valid or potentially fraudulent. The listed data is name, address, email address, phone number and IP address.

Supporting quotations

| Categories of Personal Data • Name • Address • Email address • Phone number • IP Address | Mastercard Identity Looks for patterns in the submitted information combined with signals provided by the Mastercard Identity Network to determine if the applicant is valid or potentially fraudulent. |

Qualifications & exceptions

• The assessment combines the submitted data with Mastercard Identity Network signals.
• The output describes the applicant as valid or potentially fraudulent; it does not state a final Customer decision.

Not stated here

• The notice does not describe the Mastercard Identity Network signals or how the pattern assessment is weighted.

Inference & Profiling · Context

AI Transparency Notice covers automated processing, model governance and human oversight

For additional information about automated processing, machine learning, artificial intelligence, model governance and human oversight within the Services, the notice refers readers to the Entrust Identity Verification Services AI Transparency Notice.

Supporting quotations

For additional information about the use of automated processing, machine learning, artificial intelligence, model governance, and human oversight within the Services, please refer to the Entrust Identity Verification Services AI Transparency Notice.

Qualifications & exceptions

• The notice cross-refers rather than reproducing the AI Transparency Notice's content.

Not stated here

• The AI Transparency Notice is not part of the saved source, so its governance and oversight details are not established here.

Inference & Profiling · Company practice

Each Report gives an overall result of Clear or Consider, with reasons for a Consider result

After Entrust verifies an identity or runs a check, it shares the results with the Customer in a Report. Each Report contains an overall result of 'Clear' or 'Consider'. Entrust notifies the Customer that checks are Clear if it was able to verify the User's identity and the requested checks do not show signs of fraud or other anomalies. It returns a Consider result if it was unable to verify the identity, the User was unable to pass all requested checks, or the checks show signs of fraud or other anomalies. A Consider result includes a detailed breakdown of the reasons, generated from the machine-learning models and/or human-powered processes used to verify an identity or perform a check.

Supporting quotations

Reports

Once Entrust has verified an identity or run a check, we share the results with the Customer in a report (“Report”). Each Report contains an overall result of ‘Clear’ or ‘Consider’:

If we’re able to verify the identity of a User, and the requested checks do not show signs of fraud or other anomalies, we notify the Customer that the checks are Clear.

If we’re unable to verify the identity of a User, or the User is unable to pass all requested checks or the checks show signs of fraud or other anomalies, we return a Consider result. If we return a Consider result, we will also provide a detailed breakdown of the reasons why. (The reasons are generated from the different machine learning models and/or human powered processes that are used to verify an identity or perform a check.)

Qualifications & exceptions

• Clear requires both successful identity verification and no signs of fraud or other anomalies in the requested checks.
• A Consider result can arise from inability to verify, failure to pass all checks, or signs of fraud or anomalies.
• The reasons for Consider may come from machine-learning models and/or human-powered processes.

Cross-references

• qualifies (model-proposed): Reports do not determine any outcome for a User; the Customer is solely responsible for acting on them — A Clear or Consider result is an assessment, not a decision about the User: the Customer remains solely responsible for how to act on the Report.

Inference & Profiling · Company practice

Automated processing may support or make verification determinations; Customers decide how outputs are used

The Services incorporate automated processing technologies, including machine-learning models, rules-based systems and other analytical tools, to assess identity information, detect fraud and generate verification outcomes. These technologies may be used to support or, in certain cases, make verification determinations as part of the verification process. The outputs, including Reports and related indicators, are provided to Customers. Customers determine how verification outcomes, Reports and related indicators are used within their own onboarding, authentication, fraud-prevention, compliance, account-access or other business processes.

Supporting quotations

Automated Processing and Verification Outcomes

The Services incorporate automated processing technologies, including machine learning models, rules-based systems, and other analytical tools, to assess identity information, detect fraud, and generate verification outcomes. These technologies may be used to support or, in certain cases, make verification determinations as part of the verification process. The outputs of these processes, including Reports and related indicators, are provided to Customers. Customers determine how verification outcomes, Reports, and related indicators are used within their own onboarding, authentication, fraud prevention, compliance, account access, or other business processes.

Qualifications & exceptions

• Automated technologies 'may' support determinations and, 'in certain cases', may make them.
• The Customer decides how verification outcomes, Reports and related indicators are used in its own processes.
• The listed Customer processes are examples, not an exhaustive list.

Not stated here

• The notice does not specify which verification determinations are made solely by automated processing and which require human involvement.
• The linked AI Transparency Notice is not part of the saved source.

Inference & Profiling · Company practice

Development includes training ML models and human analysts and re-running checks for quality

Developing and improving the Services includes building and improving Entrust's technology, such as machine-learning technologies and algorithms, and developing and testing new checks, products and services to better verify identity and/or detect fraud. Examples include training models to recognize a novel fraud attack or a new version of an identity document, or to minimize bias and improve performance. Entrust trains its technology to recognize patterns and make predictions about new information (machine learning), and also trains human analysts to perform those tasks when models are not best suited or are still learning. It may re-run and re-submit checks to ensure the Services work properly, particularly when testing a new feature or service for quality checks.

Supporting quotations

Developing and improving the Services includes building and improving our technology (such as our machine learning technologies and algorithms) and developing and testing new checks, products and services to better verify a User’s identity and/or detect fraud. For example, we may need to train our models to recognize a novel fraud attack, a new version of an ID document or to minimize bias and improve performance. As part of this work, we train our technology to recognize specific patterns in information and make predictions about new sets of information based on those patterns. This is known as machine learning. We also train our human analysts to perform those tasks so they can assist when our machine learning models aren’t best suited for the task or are still learning. Sometimes, we’ll also re-run and re-submit checks to ensure the Services are working properly, particularly when testing a new feature or service for quality checks.

Qualifications & exceptions

• The examples are illustrative of development work, not an exhaustive list.
• Re-running and re-submitting checks is described as sometimes occurring, particularly for testing and quality checks.

Not stated here

• The notice does not state which User data is excluded from model training or how training data is selected.

Cross-references

• defined by (model-proposed): Entrust uses Service data to improve and develop the Services unless applicable law prohibits it — The development-activities card defines what improving and developing the Services includes: technology and model building, analyst training, and quality re-runs.

Inference & Profiling · Company practice

Entrust creates statistics and benchmarking from check results, document type and device metadata

Entrust uses personal data such as check results, identity document type and device metadata to create statistics about the use and performance of the Services. It provides these statistics to Customers so that they and Entrust may use them for benchmarking and analytics, for example comparing a Customer's missed fraud rates with an industry's general missed fraud rate or analyzing fraud trends. Entrust also uses the statistics for market insights and to inform product-improvement decisions. It says an individual cannot be identified from these statistics, and where possible it pseudonymizes, aggregates and/or de-identifies the personal information, and where feasible anonymizes it.

Supporting quotations

To compile statistics, benchmarking and analytics

We use personal data (such as check results, identity document type, and device metadata) to create statistics regarding the use and performance of the Services. We provide these statistics to our Customers so that they and we may use them to, for example, conduct benchmarking and analytics (for example, to compare a Customer’s missed fraud rates to the general missed fraud rate for an industry, or to analyze fraud trends). We also use these statistics to gain market insights (for example, to enhance understanding of missed fraud trends and identify fraud patterns) and to inform our decision making on product improvements. An individual cannot be identified from these statistics and where possible, we pseudonymize, aggregate and/or de-identify (and where feasible anonymize) personal information which is used for these purposes.

Qualifications & exceptions

• The listed data (check results, document type, device metadata) is illustrative.
• Anonymization is described as 'where feasible'; pseudonymization, aggregation and de-identification are described as 'where possible'.
• The statistics are provided to Customers as well as used internally.

Not stated here

• The notice does not describe the de-identification or anonymization standards applied.