Fraudsters move victims across platforms, phone networks and banks in a single scam, yet each organisation in that chain usually sees only its own slice of the attack. A romance or investment scam may start with a social media advert, move to a messaging app, use a mobile number on a UK network and end with a payment to a mule account at a second bank. No one firm can stop that on its own data. This guide explains how fraud data sharing and consortium intelligence work in practice: the models available, the UK and EU legal bases that allow sharing, how to design what you share and how you act on what you receive. It is written for fraud, financial crime, data protection and security leads who are joining a consortium, setting up bilateral sharing, or being asked by a regulator what they share and why. For how fraud and cyber teams work together inside one organisation, see the Fraud and Cyber Convergence Guide.
Key takeaways
- Sharing is now expected, not optional. The UK APP fraud reimbursement regime splits the cost 50:50 between the sending and receiving payment firm, and the Online Fraud Charter commits large tech platforms to intelligence sharing, so every party in the chain has a reason to share.
- The legal basis is solvable but must be written down. GDPR legitimate interests, UK Data Protection Act 2018 Schedule 1 paragraph 14 for anti-fraud organisations, and section 188 of the Economic Crime and Corporate Transparency Act 2023 each cover different parts of the problem.
- Share confirmed facts, not hunches. Consortium records such as the Cifas National Fraud Database are built on confirmed conduct with a standard of proof; loose suspicion lists create unfair outcomes and complaints.
- Received intelligence is a signal, not a verdict. A match should raise risk and trigger checks, not automatically decline a customer.
- Measure the loop. Track how many shared indicators led to prevented losses, how fast you act on incoming alerts and how many matches turned out to be false.
Why cross-sector sharing matters for APP fraud
Authorised push payment (APP) fraud is the clearest example of a crime that crosses sectors. The victim is tricked into sending money themselves, so the bank sees a correctly authenticated payment. The useful evidence sits elsewhere: the advert or fake profile on a platform, the spoofed or newly issued phone number, the newly opened receiving account and the pattern of onward transfers. See the Authorised Push Payment (APP) Fraud Guide for the attack itself.
In the UK, the Payment Systems Regulator's reimbursement requirement applies to Faster Payments and CHAPS payments made on or after 7 October 2024. Eligible victims can claim up to £85,000 within 13 months of the payment, and the sending and receiving firms share the cost equally. The PSR says it supports and encourages firms to share intelligence to spot and stop fraudulent transactions. The practical effect is that receiving banks now carry half the loss for mule accounts they host, which turns inbound intelligence about mules into money saved. See the Money Mule Detection Guide.
Tech and telecoms firms have their own commitments. The Home Office Online Fraud Charter, published on 30 November 2023 and signed by firms including Amazon, eBay, Facebook, Google, Instagram, LinkedIn, Match Group, Microsoft, Snap, TikTok, X and YouTube, has nine commitment areas. Commitment 6, intelligence sharing, asks signatories to engage with initiatives that share fraud information quickly, explore internal and external data that could help prevent fraud, map data-sharing mechanisms and legal frameworks including industry databases, and agree common typologies and taxonomies. Stop Scams UK is a cross-sector membership body of banks, telecoms operators and tech firms that runs voluntary data-sharing work and the 159 service that connects callers to their bank.
Sharing models compared
Most mature programmes use more than one model. The choice depends on how fast the data must move, how sensitive it is and whether the recipient needs to act in real time.
| Model | How it works | Best for | Main risks |
|---|---|---|---|
| Consortium fraud database | Members record confirmed fraudulent conduct against a shared taxonomy; other members search at application or event time, often by API | Application fraud, identity fraud, account takeover, mule accounts | Inaccurate or stale records, weak dispute handling, members using matches as automatic declines |
| Bilateral or request-based sharing | One firm asks another for information about a specific customer or payment, or warns it after taking action | Investigating a specific APP case, tracing funds to a receiving bank | Slow, manual, inconsistent legal reasoning between pairs of firms |
| Indicator exchange (hashes, numbers, URLs, accounts) | Members share technical indicators such as phone numbers, domains, wallet addresses, device identifiers or sort code and account numbers | Blocking scam domains, adverts and numbers; platform takedowns | Indicators age quickly; personal data may still be involved (a phone number is personal data) |
| Typology and trend sharing | Aggregated or anonymised patterns, playbooks and taxonomies shared through forums or industry bodies | Cross-sector learning, horizon scanning, regulator engagement | Too slow or too general to stop live attacks |
| Privacy-enhancing collaborative analytics | Pseudonymised or encrypted matching, network analysis across institutions without pooling raw data | Finding mule networks that span banks | Complex governance; still personal data under GDPR if re-identification is possible |
How a consortium database works: the NFD example
The Cifas National Fraud Database is the best known UK example. Members come from banking, building societies, telecoms, retail, lending, insurance, gaming, mortgages and the public sector. Members record confirmed fraudulent conduct against their own organisation using fixed case types, including facility (account) takeover, identity fraud, false applications, false insurance claims, asset conversion and misuse of facility, which includes money muling. Other members search the data in real time, through a direct API connection, through third parties such as credit reference agencies, or through the Cifas portal. Cifas requires staff who use the system to complete formal training, and individuals can make a subject access request to find out what is held about them. The design lessons apply to any consortium: a fixed taxonomy, a written standard for what may be recorded, trained users, and a route for the people recorded to challenge the record.
Legal bases for sharing fraud data
Most fraud data is personal data. Account numbers, phone numbers, device identifiers linked to a customer and wallet addresses tied to an account are all in scope. The legal analysis has two layers: data protection law (may we process and disclose this?) and confidentiality and civil liability (does the customer relationship or a contract stop us?).
| Basis | Jurisdiction | What it covers | What it does not do |
|---|---|---|---|
| Legitimate interests, GDPR Art. 6(1)(f) and Recital 47 | EU and UK GDPR | Recital 47 says processing strictly necessary to prevent fraud is a legitimate interest of the controller, so it is the usual lawful basis for private-sector sharing | Still needs a documented balancing test and must be strictly necessary; does not override the data subject's right to object without assessment |
| Criminal offence data, GDPR Art. 10 | EU and UK GDPR | Allegations or findings of fraud can be criminal offence data, which needs an additional legal condition | Art. 10 itself does not give that condition; national law must |
| DPA 2018 Schedule 1 paragraph 14, preventing fraud | UK | Condition for disclosure by a member of an anti-fraud organisation, or under its arrangements, where necessary to prevent fraud; also covers processing of data so disclosed and (since 5 February 2026, inserted by the Data (Use and Access) Act 2025) processing in preparation for disclosure | Only applies through an anti-fraud organisation as defined in section 68 of the Serious Crime Act 2007; bilateral sharing outside one needs another condition |
| ECCTA 2023 section 188, direct disclosures | UK | Protects regulated-sector businesses from breach of confidence and civil liability when they share customer information with another regulated-sector business on request, or as a warning after taking a safeguarding action such as closing an account, to help prevent economic crime. In force since 15 January 2024 | Section 188(11) states it does not authorise any disclosure that would contravene data protection legislation, so you still need a GDPR basis |
| Contracts and consortium rules | Any | Membership agreements set the standard of proof, permitted uses, retention and dispute handling | Cannot create a lawful basis that data protection law does not allow |
Two practical points follow. First, section 188 solves the confidentiality problem but not the data protection one, so firms using it still need a legitimate interests assessment and, where fraud allegations are involved, a Schedule 1 condition. Second, the ICO's data sharing code of practice is a statutory code under section 121 of the Data Protection Act 2018. It recommends a data protection impact assessment for sharing that carries risk and calls a data sharing agreement good practice. The ICO notes the code is under review following the Data (Use and Access) Act, so check the current version before relying on specific wording. This guide summarises legal provisions for security and fraud teams and is not legal advice. See the DPIA Guide and the GDPR Compliance Guide for Security Teams.
Designing what you share
The quality of a sharing programme is set by the record standard, not by the technology. Agree these elements before the first record is exchanged.
- Standard of proof: Define when a case is confirmed enough to record. A common test is that there is enough evidence that the conduct occurred and that it would be reasonable to report it to the police, rather than a single model score.
- Taxonomy: Use fixed case types (account takeover, false application, mule activity, impersonation) and fixed role labels (victim, perpetrator, facilitator, unknown). Never record a victim as a subject of fraud.
- Data minimisation: Share the identifiers the recipient needs to act: for a mule account, sort code and account number, opening date and case type. Do not share full case notes by default.
- Retention: Set a fixed retention period per case type and delete or archive on schedule. Stale records drive false matches.
- Permitted use: State whether matches may be used only for fraud prevention, or also for credit or onboarding decisions. Uses beyond fraud prevention need their own basis.
- Transparency: Tell customers at onboarding that data may be shared with fraud prevention agencies and what that means. A layered fair processing notice is the norm.
- Corrections and disputes: Provide a route for an individual or member to challenge a record, with a deadline to investigate and a duty to amend or remove inaccurate data across the consortium.
Indicator types and their sensitivity
| Indicator | Typical source | Useful lifetime | Personal data? |
|---|---|---|---|
| Receiving account (sort code and account number, IBAN) | Bank fraud teams | Days to weeks for mules; longer for repeat offenders | Yes |
| Phone number used for scam calls or SMS | Telcos, banks, consumer reports | Hours to days | Usually yes |
| Scam domain or URL | Platforms, banks, takedown providers | Hours to days | Usually no, unless it names a person |
| Advert or profile identifier | Social media, marketplaces, dating services | Days | Possibly |
| Device or session fingerprint | Banks, merchants, fraud platforms | Weeks | Yes when linked to a person |
| Crypto wallet address | Exchanges, analytics firms, banks | Long, but funds move quickly | Yes when linked to a customer |
| Typology description | Fraud teams, industry bodies | Months | No, if properly anonymised |
Acting on received intelligence
A consortium match or an inbound warning is evidence about another organisation's experience, not proof about your customer. Treat it as a risk signal that changes the next step.
- Route by case type: A match on a mule account at onboarding goes to the financial crime queue; a match on a phone number in a payment journey triggers a stronger scam warning and a call-back.
- Combine with your own signals: Feed matches into your rules engine or score as features, not as overrides. See the Fraud Prevention Guide for Digital Businesses for scoring design.
- Apply proportionate actions: Step-up verification, holding a payment for review, restricting outbound faster payments or requesting documents are usually more proportionate than refusal.
- Record your own decision: Document why you acted, so that a complaint or ombudsman review can see a reasoned human decision rather than a blind reliance on a third-party flag.
- Close the loop: Tell the sharer whether the intelligence was confirmed, so the consortium learns which members and indicators are reliable.
How to implement a sharing programme
- Map your fraud journeys: For your top three fraud types, list which external parties hold evidence you lack (platforms, telcos, other banks) and which of your data would help them.
- Choose the vehicles: Join an existing consortium or anti-fraud organisation before building bilateral links. Existing schemes come with tested legal bases, taxonomies and dispute processes.
- Complete the legal work: Legitimate interests assessment, Schedule 1 condition where criminal offence data is involved, DPIA, data sharing agreement and an updated fair processing notice. Involve the DPO from the start.
- Build the integration: Connect by API where real-time decisions are needed; restrict access to trained, named users; log every search and every record created.
- Pilot in shadow mode: Run searches and record what you would have done for four to eight weeks without acting, then compare against confirmed outcomes to measure false matches.
- Go live with proportionate actions: Start with step-up and review actions; move to holds or restrictions only for case types with proven precision.
- Govern and review: Quarterly review of record quality, disputes, retention and use; annual review of the legal basis. Roll back by suspending automated actions on a case type while keeping manual searches.
Small and large organisations: a small lender or fintech gains most by joining one established consortium and using its portal or a provider integration; building bilateral agreements is rarely worth the legal cost. Large banks and platforms typically run several channels at once: consortium databases, bilateral request handling under section 188, indicator feeds to and from platforms, and typology forums.
Where identity fits
Sharing platforms are high-value targets. A consortium API key or service account that can search and record fraud markers could be abused to look up individuals or to plant false records against them. Treat these credentials as privileged non-human identities: store them in a secrets manager, scope them to search or record functions separately, rotate them, restrict them by source IP or mutual TLS where the provider supports it, and alert on unusual search volumes. User access to the portal should be named, trained and reviewed at least quarterly. See the Identity Fraud Prevention Guide for the customer side of identity checks.
Detection and measurement
Log every outbound record (who created it, case type, evidence reference), every search and match, and every action taken on a match. Monitor for searches without a linked application or case, unusually high record creation by one user, and records later withdrawn on dispute.
- Prevented loss attributable to shared data: value of payments stopped or applications declined where a shared indicator was a deciding factor, confirmed after review.
- Match precision: share of matches that your own investigation confirmed as fraud, by case type and by source.
- Time to act on inbound alerts: median time from receiving a warning about a receiving account to restricting it. For mule accounts this should be measured in hours.
- Contribution rate: confirmed cases recorded to the consortium as a share of confirmed internal cases; members that only search weaken the scheme.
- Dispute outcomes: number of records challenged and the share amended or removed, which shows whether your standard of proof is working.
Framework mapping
| Framework | Reference | How it applies |
|---|---|---|
| NIST CSF 2.0 | ID.RA-02 | Threat intelligence is received from information sharing forums and sources |
| NIST CSF 2.0 | RS.CO-03 | Information is shared with designated internal and external stakeholders |
| NIST CSF 2.0 | GV.SC-07 | Risks from third parties, including consortium operators and data providers, are assessed and monitored |
| NIST SP 800-53 Rev. 5 | AC-21 Information Sharing; PM-16 Threat Awareness Program; PT-2 Authority to Process PII | Sharing decisions, cross-organisation threat awareness, and a documented authority to process the personal data involved |
| ISO/IEC 27001:2022 Annex A | 5.7 Threat intelligence; 5.14 Information transfer; 5.34 Privacy and protection of PII | Collecting and acting on intelligence, agreements for transfer, and privacy obligations |
| UK and EU GDPR | Art. 6(1)(f), Art. 10, Art. 35 | Lawful basis, criminal offence data, DPIA |
Common failure patterns
- Treating a match as a decision: automatically declining or closing accounts on a consortium hit, which harms innocent people with similar details and fails a complaint review.
- Recording suspicion, not fact: loading model alerts or unconfirmed cases, which lowers precision for every member and creates data protection exposure.
- Relying on section 188 alone: assuming the confidentiality protection also provides a data protection basis, when the Act says it does not.
- Search-only membership: taking intelligence without contributing confirmed cases, so the scheme never sees your fraud.
- Slow inbound handling: warnings about receiving accounts sit in a shared mailbox for days, while the mule empties the account within hours.
- Victims recorded as subjects: poor role labelling marks scam victims as fraud subjects, blocking them from banking and compounding the harm.
Practitioner checklist
- List the external parties that hold evidence for your top fraud types and the channel you use for each.
- Document the lawful basis per sharing flow: legitimate interests assessment, Schedule 1 condition where needed, and section 188 reliance where it applies.
- Complete a DPIA and a data sharing agreement for every new flow, and update customer privacy notices.
- Write a standard of proof and taxonomy for outbound records, with victim and perpetrator roles kept separate.
- Configure matches as risk features with proportionate actions, never automatic refusals.
- Set a target time for acting on inbound mule account warnings and measure it.
- Store consortium API credentials in a secrets manager with scoped permissions and alerting on search volume.
- Run a dispute process with deadlines and propagate corrections to the consortium.
- Report prevented loss, match precision, contribution rate and dispute outcomes quarterly.
Standards and references
- UK legislation: Economic Crime and Corporate Transparency Act 2023, section 188
- UK legislation: Data Protection Act 2018, Schedule 1 paragraph 14 (preventing fraud)
- ICO: Data sharing code of practice
- Home Office: Online Fraud Charter
- Payment Systems Regulator: APP fraud reimbursement protections
- Cifas: National Fraud Database
- Stop Scams UK: Cross-sector collaboration against scams
- NIST: Cybersecurity Framework 2.0
Related NHI Mgmt Group resources: Fraud and Cyber Convergence Guide · Authorised Push Payment (APP) Fraud Guide · Money Mule Detection Guide · Threat Intelligence Guide. This guide summarises UK and EU legal provisions on fraud data sharing and is not legal advice.