Publisher disclosure: Truthlocks is the publisher and a product being compared. Researched and written with AI assistance; reviewed against the primary sources linked below on October 28, 2026. This is a company-authored architectural assessment, not an independent product test.
A bank statement can look ordinary and still contain a fabricated balance. Another statement can look unusual because a legitimate banking system exported it differently. Comparing Resistant AI with Truthlocks starts with this distinction: detecting suspicious documents and verifying issuer-signed evidence answer different questions. A lender receiving arbitrary uploads needs a different first control from a university checking records it issued itself.
What Resistant AI does
Resistant AI markets document fraud detection for fake, modified and AI-generated documents, alongside transaction-monitoring products. Its Document Forensics overview describes layered analysis of submitted files. This makes it relevant when the sender supplies a PDF or image without a trustworthy signature. These are vendor-described capabilities, not a comparative benchmark we independently ran.
The buying question is not whether a detector can produce a score. It is whether its findings help your reviewers distinguish manipulation from ordinary document processing in your own population. Country coverage, scans, screenshots, unfamiliar templates and generative editing all belong in that evaluation.
What Truthlocks verifies
Truthlocks provides signed records, issuer and key references, verification results and documented proof-bundle interfaces. The public evidence page separates dated first-party API checks from independent assurance. The public signature exercise demonstrates a valid synthetic receipt and deliberate tampering failures. Neither constitutes a document-forensics accuracy study.
Its deepfake guide explicitly states that the endpoint scores supplied indicators or external scores; it does not fetch and inspect the referenced image, video or document. Treat that interface as risk-signal processing, not as an alternative image detector. A low score cannot establish authenticity when no useful evidence was supplied.
Compare the evidence each approach needs
| Criterion | Resistant AI | Truthlocks |
|---|---|---|
| Starting input | Submitted documents for forensic assessment | A signed record and a trustworthy route to its issuer key |
| Main question | Does this document show fraud indicators? | Does this record match the signed evidence? |
| Unknown provenance | Relevant to investigation of untrusted uploads | A missing trusted signature cannot be repaired by hashing the upload |
| Decision boundary | Signals interpreted through review policy | Signature, status and issuer checks interpreted through acceptance policy |
| Pilot evidence | False positives, misses and review workload | Alteration, wrong-key, revocation and proof checks |
Where Resistant AI may be stronger
For a queue of bank statements from unknown sources, forensic detection addresses the immediate evidence gap. Requiring every applicant's bank to issue a new cryptographic record may be operationally unrealistic. A detector can help prioritize investigation without that prerequisite. This is an architectural advantage for the intake problem, not proof that any detector catches every forgery.
False negatives matter: a sophisticated forgery may escape a chosen threshold. False positives matter too: a legitimate re-export or scan can create unusual artifacts. Request reason codes and assess representative files rather than accepting an aggregate accuracy claim. Preserve uncertain cases instead of silently converting an inconclusive result into approval.
Where Truthlocks is a better architectural fit
When an authoritative issuer already participates, the requirement may be to check exactly what that issuer signed and whether the record remains acceptable. Truthlocks is a candidate for this narrower contract. For example, an employer can check an institution's credential reference and detect a changed qualification field. The institution's authority still needs validation; an attacker can sign false statements using a key they control.
The useful distinction is between the integrity of a statement and confidence in its source. Maintain an issuer admission process, record who approved each issuer, and define what happens when a signing key is compromised. A mathematically valid signature from an unapproved issuer should not inherit institutional authority.
Example: a lending intake with two evidence paths
Consider a hypothetical lender receiving a scanned statement and an employer-issued income credential. Route the scan to document analysis and request clarification for suspicious edits. Independently verify the signed credential against the employer's accepted key and current status. Reconcile the figures with the lending application. A clean scan and a valid credential can still disagree; that disagreement needs a documented human or policy decision.
If you retain the decision as a signed record, include the evidence versions, review time and policy version. Do not embed unnecessary bank-account or identity data in public records. Later reviewers should be able to reconstruct why the loan was considered, without receiving unrestricted access to the original applicant file.
Integration and pilot acceptance
Test an untouched document, a legitimate format conversion, a changed balance, an unknown issuer, an expired credential and a signing-key rotation. Score detection outcomes separately from signature outcomes. Measure reviewer workload as well as technical success. An HTTP 200 response says the service answered; the result fields determine whether your rule passed.
For Truthlocks, exercise the verification workflow in a scoped sandbox and inspect transparency inclusion separately from signature validity. For either service, agree retention, deletion, incident notification and data-access terms before sending customer documents. This comparison does not establish pricing equivalence, certifications or production-scale reliability.
Which should you choose?
Prioritize Resistant AI evaluation when your problem is suspicious incoming documents without authoritative provenance. Evaluate Truthlocks when the problem is issuing and checking signed institutional or application records. Combine the approaches when both gaps exist, but budget for the integration and the reviewer decision between them.
Buyer FAQ
Can Truthlocks detect an AI-generated payslip by itself?
The documented supplied-signal endpoint is not an independent payslip image detector. A signature check is useful only when there is trustworthy signed evidence to compare.
Does a valid signature eliminate fraud?
No. An issuer can make an error, a key can be misused, or a truthful record can be presented outside its intended context.
What should an alternatives evaluation measure?
Separate forensic detection quality, issuer trust, record integrity and operational review costs. A single combined pass rate hides different failure modes.
Explore the comparison collection · Explore Truthlocks for your workflow. Validate required capabilities in a scoped pilot before adopting either product.