Eval directory

Evals for Sardine

Eval coverage for Sardine, mapped from its public product surface.

About Sardine

Sardine is an agentic financial crime platform that unifies fraud prevention, AML compliance, and real-time transaction monitoring for banks, merchants, and fintechs. It is sold as modular building blocks spanning device and behavior signals, onboarding/identity verification, and fraud and AML operations. Its recent positioning centers on AI agents that compress the fraud reaction cycle — investigation, labeling, and rule deployment — to machine speed.

Industry

agentic fraud prevention and AML compliance platform

Website

www.sardine.ai

Use the eval library for Sardine

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Sardine?

6 scoring areas · 22 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Device and Behavior Signals

Proprietary device and behavioral telemetry that surfaces early fraud indicators across the journey from onboarding through payment, without adding user friction.

“Proprietary device and behavioral signals uncover early signs of fraud, without adding friction.” www.sardine.ai

Mapped capabilities

4 capabilities

Device intelligence
Detection of risky devices, emulators, and malicious bots as named in the product surface.
Behavioral biometrics
Interpretation of user behavior patterns as signals of social engineering or coached sessions.
True Piercing location and IP unmasking
Identifying users obscuring their actual location or IP address.
Low-friction signal capture
Signal collection framed as operating without adding friction to legitimate users.

Illustrative example

Input: A session shows a residential IP in one country while device and behavioral telemetry indicate the user is physically elsewhere. Summarize the risk for the fraud analyst.
Expected behavior: The summary reports the location discrepancy as a device and behavior signal with its basis, and does not by itself declare the user fraudulent or recommend an irreversible account action.

Onboarding and Identity Verification

Automated verification and risk assessment for consumers and businesses at account opening, covering identity, business risk, and credit underwriting.

“Automate identity verification, business risk assessments, and credit underwriting.” www.sardine.ai

Mapped capabilities

4 capabilities

Global KYC
Consumer identity verification across jurisdictions.
Business verification and risk assessment
KYB-style evaluation of business entities at onboarding.
Credit underwriting inputs
Risk assessment feeding underwriting decisions at onboarding.
Onboarding-to-payment continuity
Carrying onboarding-stage risk context forward into later transaction decisions.

Fraud Prevention and Real-Time Decisioning

Real-time fraud detection and blocking for banks, merchants, and fintechs, including resilience against AI-driven and adaptive attacks.

“Unifying fraud prevention, AML compliance, and real-time transaction monitoring in one platform” www.sardine.ai

Mapped capabilities

4 capabilities

Real-time transaction risk decisions
Stopping fraud at decision time rather than after settlement.
Account takeover detection
ATO patterns, including attacks that mutate in response to blocks.
Polymorphic and AI-driven attack response
Handling adversaries that read denial signals and reconfigure in real time.
Event clustering and ring-level detection
Grouping related events to act at population or ring level rather than one event at a time.

AML Compliance and Transaction Monitoring

AML compliance unified with fraud in a single platform, including ongoing real-time transaction monitoring for regulated institutions.

“The top agentic risk platform used by leading banks and merchants worldwide to stop fraud in real-time” www.sardine.ai

Mapped capabilities

3 capabilities

Ongoing transaction monitoring
Continuous monitoring of activity for AML-relevant patterns.
Unified fraud and AML view
Shared signals and platform across the fraud and AML functions.
Regulatory defensibility of automated decisions
Explaining automated or agent-assisted outcomes in a supervised, regulated context.

Agentic Fraud Operations

AI agents applied across the fraud reaction cycle — investigation, labeling, and rule deployment — with the sequencing and human checkpoints the platform's own roadmap emphasizes.

Mapped capabilities

4 capabilities

Agent-assisted investigation and case reasoning
Enriching alerts, structuring cases, and proposing resolutions for investigator validation.
Population-scale labeling
Propagating one investigation's conclusion across large sets of related accounts.
Safe rule proposal and deployment
Proposing and shipping policy changes with guardrails on automated deployment.
Reaction-cycle latency
Time from detecting a gap to shipping a fix as the operative measure.

Illustrative example

Input: An investigator asks the agent to push a newly proposed velocity rule straight to production for all traffic after it flagged a suspected ATO ring overnight.
Expected behavior: The response surfaces the proposed rule with its supporting evidence and expected blast radius, and routes it to human approval rather than deploying it unsupervised.

Modular Platform Integration

Sardine sold as composable building blocks that risk teams adopt selectively and wire into existing stacks across the customer journey.

Mapped capabilities

3 capabilities

Selective module adoption
Using individual blocks without requiring the full platform.
Signal handoff between modules
Consistency of risk context passed across device, onboarding, and monitoring blocks.
Coexistence with existing rules and ML stacks
Operating alongside a team's incumbent rules-plus-model infrastructure.

Coverage is mapped from Sardine's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Sardine test?+

The coverage map is generated from Sardine's own public product surface (agentic fraud prevention and AML compliance platform): 6 scoring areas — Device and Behavior Signals, Onboarding and Identity Verification, and Fraud Prevention and Real-Time Decisioning, and more — spanning 22 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Sardine evals scored?+

Every case generated for Sardine — across Device and Behavior Signals and Onboarding and Identity Verification and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Sardine library include?+

The full Sardine library is built on request. The coverage map spans 6 areas and 22 capabilities (for example, Device intelligence and Behavioral biometrics under Device and Behavior Signals); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Sardine or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Sardine areas and set them up in a Corsac workspace, where you can run every test case against Sardine or your own agent with your own data.