Verify before you act.
Got a message that asks you to click, pay, or share a code? VERITAS shows you the evidence, what it can and cannot verify, and the safest next step, instead of just guessing “scam or not”.
Links in your message are never opened. Raw message text is not stored.
chase.com yourself.How it works
- 01Submit
Paste a message, drop a screenshot, or enter a link.
- 02Extract
Entities, claims, links, deadlines and what is being asked of you.
- 03Check
Social-engineering rules, URL/domain signals and cross-checks run in parallel.
- 04Explain
Risk and confidence kept separate, every reason labelled by source.
- 05Verify before you act
A safe, independent route to confirm. Never the link in the message.
Each finding says where it came from: the message itself, a fixed rule, an outside source, or an AI inference.
When something can’t be checked, VERITAS says “Unable to independently verify” and tells you how to.
The output is a plan: what to do now, what not to do, and who to report to.
Every conclusion shows its work
Four kinds of evidence, labelled everywhere you look. The AI can add concern, but only inspectable checks decide the verdict.
Text or data literally present in what you submitted.
A fixed, inspectable security rule produced this. Same input, same result.
Retrieved from an outside source at analysis time, with timestamp.
An AI model's judgement. Validated, bounded, and never able to change the verdict level on its own.
Not sure about a message?
Check it in seconds, then decide with evidence in front of you.
Open the analyzer