How does Plaid use AI for fraud and risk?
Plaid's fraud and risk AI runs on signals from its account-connection network and comes in four products: Signal for ACH payment risk, Protect and its Trust Index for user fraud, Beacon for shared fraud reports, and Identity Verification. Signal, a machine learning risk engine whose availability Plaid expanded on November 17, 2022, scores each ACH payment for return risk; Signal 4, announced October 6, 2026, caught 126% more risky dollars than v3 at a 5% decline rate in testing. Protect launched in beta on June 12, 2025 with the first Trust Index model, drawing on fraud signals across a billion devices; Ti2 caught 30% more fraud (October 2025) and Ti3 up to 41% more at the same false-positive rate in testing (May 2026). In October 2026 Plaid added a transformer-based Fraud Foundation Model that showed a 40% improvement in internal tests. Beacon, launched June 22, 2023, lets members report fraud and screen identities against each other's reports. Plaid's 2025 review says Identity Verification caught 47% more fraud than a leading identity provider in head-to-head testing. These are Plaid's own results; no independent evaluation is published.[1][2][3][4][5][6][7][8]
Data as of Oct 19, 2026 · Plaid profile updated Oct 19, 2026 · 21 sources cited on this page · 52 fintechs profiled on this site · Markdown
How has Plaid's ACH risk model developed?
| Version or milestone | What Plaid reported | Date |
|---|---|---|
| Signal announced | American Banker reported that Plaid had announced Signal the previous fall and that it used close to a thousand risk factors, scoring each transaction as money moves | May 18, 2022[9] |
| Wider availability | Over 1,000 risk factors, more than 60 attributes; an investment platform gave 96% of new users accelerated access to funds with nearly no increase in returns; a digital wallet cut return losses 43% while stepping up under 1% of users | November 17, 2022[1] |
| Payment risk platform | More than $185 billion decisioned; core-model recall up over 140%; no-code rules, balance checks or ML scores; losses from ACH returns cut by up to 80% for clients like Uphold | October 1, 2025[10] |
| Signal 4 | 126% more risky dollars caught than v3 at a 5% decline rate; 7% more good dollars approved at a 1% return rate; sequential-model features make up 8 of the top 20 features; AI reads raw transaction descriptions to find returns never reported to Plaid; rolls out broadly in early Q4 2026 | October 6, 2026[2] |
| Return-data feedback | Customers who share return data via API have seen a 2-3x increase in dollar recall | October 6, 2026[2] |
How does Plaid detect fraud by users and identities?
Results marked retro-analysis estimate what a model would have caught on a customer's historical data.
| Product | What it does | Reported result | Date |
|---|---|---|---|
| Beacon | Anti-fraud network: members report fraud via API or dashboard and screen identities against other members' reports without seeing who reported them | Launched with design partners Tally, Uphold, Possible Finance, Uplift, CreditGenie and Veridian Credit Union | June 22, 2023[7][11] |
| Beacon account insights | 40+ insights on a Plaid-linked account, such as account age, connected apps and identity changes | A neobank analysis showed 65x higher unauthorized-return risk when three or more browsers linked the account within a week; more than 20 participating companies | May 22, 2024[12] |
| Identity Verification | Fraud calibration labels plus 11 new machine learning-driven scores and an overall Trust Index per user | 300% month-over-month increase in customer-submitted fraud labels | October 15, 2024[13] |
| Protect, Trust Index 1 | Real-time fraud scores from 10,000 high-signal attributes across a billion devices; natural-language search in the dashboard | Stepping up 5% of users could have caught nearly 40% of first-party fraud for an early fintech customer | June 12, 2025[3] |
| Trust Index 2 | Adds bank-account transaction history and live network graph analysis, with twice the training data | 30% more fraud caught; in retro-analysis, 50% of a crypto firm's fraud losses could have been prevented | October 14, 2025[4] |
| Identity Verification upgrades | Detection of GenAI deepfakes and synthetic video via expanded liveness checks (2025); improved large language models evaluate ID documents (March 2026) | 47% more fraud caught than a leading identity provider in head-to-head testing | 2025–2026[8][14] |
| Protect out of beta | Network-level mapping across 500 million bank accounts, users and devices | Retro-analyses: detection up 40-59% while stepping up 5-10% of users; on average 46% more first-party fraud detected and 52% of fraud dollar losses prevented | March 12, 2026[15] |
| Trust Index 3 | Fraud graph grown nearly 10x, traversed up to nine hops in real time; enhanced device identification through Plaid Link | Up to 41% more fraud at the same false-positive rate in early testing | May 18, 2026[5] |
| Fraud Foundation Model | Transformer pretrained on hundreds of millions of proprietary data points, fine-tuned on customer fraud labels, feeding the Trust Index | 40% improvement in fraud detection over previous Trust Index models in internal tests | October 6, 2026[6][16] |
What limits apply to Plaid's fraud models, and what is not disclosed?
Plaid's identity products came partly by acquisition: American Banker reported in May 2022 that Plaid Identity Verification is Cognito's technology, which Plaid bought and built into Plaid Link.[9]
Signal is positioned outside credit reporting. Plaid's launch post says Signal is not a consumer report under the Fair Credit Reporting Act and may not be used to decide a consumer's eligibility for credit, insurance or employment.[1]
Beacon depends on members sharing. American Banker reported at launch that Plaid sets benchmarks for how many fraud examples each organization should generate and monitors them, and aimed to extend Beacon to its 100 largest customers.[11]
Regulation is pushing the same way: Nacha's 2026 rules make fraud monitoring before ACH origination a requirement, phased in on March 20 and June 19, 2026, and Plaid markets account-ownership checks and Signal against them.[17]
What is not disclosed: production false-positive rates, fraud dollars stopped in total, how many companies use Protect (Plaid said "dozens" in March 2026) or Beacon since 2024, and any third-party test of the Trust Index or Signal. Every result above is Plaid's own testing or retro-analysis.[15][12][5]
What does the Plaid profile record on this topic?
The rows below are the entries of the Plaid AI strategy profile that bear on this page, as of Oct 19, 2026; the profile holds the full record of 16 dated moves, 11 use cases and 26 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Trust Index (Plaid Protect) | Network fraud model; Plaid says the current Trust Index catches 30% more fraud than earlier versions, and Trust Index 3, announced May 2026, catches up to 41% more fraud in testing at the same false-positive rate. On October 6, 2026 Plaid announced a Fraud Foundation Model, a transformer pretrained on behavioral and transactional sequence data that feeds the Trust Index; Plaid says it shows a 40% improvement in fraud detection over previous Trust Index models in internal tests. | In production | [18][19][6] |
| LLM document checks in Identity Verification | Identity Verification uses improved large language models to evaluate ID documents, adding screen and printed-copy detectors (March 2026). | In production | [14] |
| Decagon AI agents with Plaid-verified accounts | Decagon's conversational AI support agents connect to and act on a consumer's bank account through Plaid inside the chat, covering tasks such as reconnecting a failed autopay, fraud resolution, insurance claims and loan verification. | Rolling out | [20] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Sep 28, 2026 | Plaid and Decagon partner on conversational AI agents | Plaid partnered with Decagon so Decagon's AI customer-support agents can connect to, verify and act on a consumer's bank account through Plaid inside the conversation, without redirecting to a separate login; example uses include reconnecting after a failed autopay, fraud resolution, insurance claims and loan-verification steps, with the agent able to resume multi-step workflows across sessions. Decagon's Roger Liang said Plaid's infrastructure lets its agents "verify a bank account" and handle a stuck transfer or complete a payment inside the chat. | [20] |
| Jun 25, 2026 | Sequential foundation model | A model that reads transaction sequences over time; Plaid reports 26.5% more dollar value in ACH returns prevented at a fixed 1% action rate and default risk cut by 13.6% at a 70% approval rate. | [21] |
| May 21, 2026 | Effects 2026: foundation models, Trust Index 3, LLM-based Income | Trust Index 3 (up to 41% more fraud caught at the same false-positive rate in testing), Plaid Income rebuilt on a proprietary transformer-based LLM (86% precision), a Cash Advance Index, Guaranteed Payments and MCP servers for AI coding agents, with an MCP server for banks promised later in the year. | [19] |
| Feb 19, 2026 | Plaid sets out its intelligence and trust layer for agents | A Plaid post, featured by CTO Will Robinson in the March product update, said Trust Index catches 30% more fraud than earlier versions and Plaid Signal has analyzed more than $185 billion in payments. | [18][14] |
What is Plaid Protect?
Plaid Protect is a real-time fraud intelligence system launched in beta on June 12, 2025 and out of beta by March 2026. It scores users with the Trust Index, a machine-learning model fed by device, bank-account and network signals; Plaid says its third version catches up to 41% more fraud at the same false-positive rate in testing.
What is the Plaid Trust Index?
The Trust Index is Plaid's fraud risk score. Ti1 launched with Protect in June 2025 on 10,000 high-signal attributes; Ti2 added transaction history and graph analysis and caught 30% more fraud (October 2025); Ti3 expanded the fraud graph nearly 10x (May 2026); and in October 2026 a Fraud Foundation Model began feeding it.
What is Plaid Beacon?
Beacon is Plaid's anti-fraud network, launched June 22, 2023. Members report fraud such as stolen identities or account takeovers, and other members screen new users against those reports without learning who filed them. In May 2024 Plaid added 40+ insights about linked bank accounts and said more than 20 companies took part.
How accurate is Plaid's ACH risk model?
Plaid reports that Signal 4, announced October 6, 2026, caught 126% more risky dollars than Signal 3 at a 5% decline rate and approved 7% more good dollars at a 1% return rate in testing. Earlier, it said core-model recall improved over 140% in a year (October 2025). These are Plaid's own tests.
Can Plaid Signal be used for credit decisions?
No, by Plaid's own terms. Its November 2022 launch post says Signal is not a consumer report under the Fair Credit Reporting Act and may not be used to determine eligibility for credit, insurance or employment.
- Announcing Plaid Signal: Unlock instant ACH and reduce return risk
- Signal 4: the strongest generation of ACH risk insights that catch 126% more risky dollars
- Introducing Plaid Protect – the next chapter in fraud prevention
- Introducing Ti2: The next generation of Plaid Protect's Trust Index
- Ti3 is here: A bigger graph for a fast moving fraud landscape
- Fraud Foundation Model: The next chapter in fraud prevention for Plaid
- Introducing Beacon
- What we shipped in 2025
- Plaid expands beyond data aggregation into ID and fraud tech
- Introducing the Signal payment risk platform: a full solution for ML-powered risk assessments
- Plaid leans on banks, fintechs to create a stolen-identity database
- Beacon adds bank account insights to help companies strengthen fraud defense
- Introducing fraud calibration and Trust Index for Plaid Identity Verification
- Product Updates - March 2026
- New era of fraud network intelligence: Early results from Plaid Protect
- Plaid Launches Credit, Fraud Models Fueled by Cash Flow Data
- 2026 Readiness Check for Nacha Fraud Monitoring Rule Changes
- Why intelligent finance needs purpose-built AI infrastructure
- Everything we announced at Plaid Effects 2026
- Plaid and Decagon partner to power more impactful financial experiences with AI agents
- Learning the grammar of money with a sequential foundation model
When Plaid moves on AI, you'll read it here first.
when a bank, fintech or regulator moves on AI, it is in the next morning's brief · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning