How Cluttersignal works

See exactly how the trust layer decides what to trust.

Cluttersignal screens what reaches you — in your inbox and in your browser — and shows its work. Here's the engine that decides, the email pipeline, the on-device browser path, and the security practices behind them.

How the trust layer decides

One engine, six questions, one recommended action.

Before anything reaches you — in your inbox or in your browser — Cluttersignal reasons through the same six dimensions, then folds them into a 0–100 Trust Score and a single recommended action.

Identity

Is this really who it claims to be? SPF, DKIM, and DMARC authentication, plus lookalike, homoglyph, and typosquat domain detection.

Relationship

Does this match your history? Prior correspondence and sender familiarity from your own mailbox.

Threats

Phishing, malware, or scam signals? Scam and phishing language, malicious-URL reputation, and attachment scanning where files are in scope.

Context

Does this request make sense? Domain registration age and reputation, and password, payment, and MFA form detection.

Privacy

What could this expose? Flags pages requesting credentials or payment and known trackers — suspicious data-collection behavior, not a full data audit.

Decision

What should you do next? One clear recommended action — verify, proceed, or walk away.

Current controls

What's actually in place today.

Specific safeguards we operate today — not certification language we haven't earned. If a control isn't on this page, we don't claim it.

Encrypted credentials

Connected account credentials are encrypted at rest and protected separately from ordinary application session handling.

Approval required

AI drafting does not mean autonomous execution. Sends, edits, and other sensitive actions pause for human approval.

Threat screening

Inbound messages are screened before they become another thing the user has to reason about in the inbox.

Auditable activity

Agent actions, approvals, and important workflow changes are visible in activity history instead of disappearing behind automation.

Cluttersignal flow

7-phase protection.

Every email passes through the Cluttersignal pipeline before you ever see it. CSS-hidden text, invisible font exploits, and prompt injection attempts are stripped at the sanitization boundary, so the AI never sees raw hostile content.

1.

MIME Parsing

Dissect headers, body, and attachments.

2.

Header Forensics

Check SPF, DKIM, and Reply-To mismatch signals.

3.

Sender Reputation

Score domain age, homoglyph risk, and freemail patterns.

4.

Threat Scoring

Combine weighted signals into a composite risk score.

5.

Prompt Injection Detection

Strip hidden instructions targeting AI before model access.

6.

AI Explainer

Generate plain-language threat analysis for human review.

7.

Sanitized Evidence

Return clean, AI-safe output with the boundary intact.

Cluttersignal is a defense-in-depth layer — not a guarantee

Cluttersignal is designed to catch the vast majority of phishing, scam, malicious-link, and prompt-injection attempts before they reach you. No automated threat-screening system catches every threat. Cluttersignal occasionally misses malicious mail and occasionally flags legitimate mail.

Treat it as one of several defenses — alongside your provider's spam controls, your own judgment, and any organization-level security tooling — not as a replacement for them. We do not warrant that Cluttersignal (or any AI-powered feature) will detect every fraudulent or unwanted message. See Terms § 6 for the full disclaimer.

On every site you open

The same engine, running on-device in your browser.

The Cluttersignal Extension brings the trust layer to the whole web. The badge is computed on your device — only the content you explicitly check is sent for a deeper look, then analyzed and discarded, not stored.

1.

On-device page signals

The badge reads form fields, links, and page structure locally in your browser to decide what to flag.

2.

Domain checks

Resolves the real domain behind the page and weighs its registration age and reputation.

3.

Lookalike detection

Catches homoglyph and typosquat domains pretending to be a brand you trust.

4.

Verdict + badge

Returns a safe / caution / suspicious / dangerous verdict as a quiet badge on the site — free, on every page.

5.

AI trust report (Signal Pro)

On request, the content you explicitly check is analyzed for the comprehensive Trust Report, then discarded — not stored.

Data handling

What we store, what we use it for, and what you control.

Three plain answers: what stays in the product, what powers the workflow, and what you can disconnect or delete on your own.

What we store

The product stores the information it needs to deliver the workflow: connected account metadata, recent message context, encrypted credentials, and the activity needed to support approvals and history.

What we use it for

Clutterstrike uses that data to organize conversations, power drafting and summarization, screen risky content, and keep tasks and follow-up tied back to source context.

What you control

Users can disconnect providers, control account connections, and manage what workflows they want active instead of being locked into opaque automation defaults.

Report an issue

Found a security problem? Tell us directly.

If you find a vulnerability or something that feels unsafe, contact the team at security@clutterstrike.com and include enough detail for us to reproduce and respond.