Private, on-device review

Pricing

See what a surveillance model could notice.

Paste public posts or inspect a downloaded archive. Get a transparent, defensive exposure review of political speech, dissent, networks, and identifying details—without connecting an account or sharing a password.

Add your material

01 / INPUT

Up to 200,000 characters0 chars

Date filtering applies to file uploads, which carry timestamps.

Public figures, parties or institutions to check OPTIONAL

Public officeholder, party, role, or institution names only. Never add private individuals, handles, or contacts. The list stays in this browser tab and never leaves your device.

Your text stays in this tab and is discarded when you close or refresh it. Files are read locally; nothing is uploaded.

02 / REVIEW

A lens, not a verdict.

This meter shows signals an authoritarian monitoring system could treat as legible—especially criticism of leaders, ruling parties, and state institutions. It cannot predict what any real system or government would do.

S-01Political figuresLeaders, parties, opposition, dissent

S-02Civic activityEvents, campaigns, public institutions

S-03Network linksHandles, groups, invitations

S-04Time & placeMeeting details and movement clues

S-05Criticism of powerAnti-regime language and directed criticism

S-06Identity cluesContact details and public handles

Exposure profile

—

0/ 100

LOWER — FEW CHECKED SIGNALS

Few machine-legible signals

This text contains few of the markers checked by this model.

0signal matches

0identity clues

Neutraltone estimate

What raised the meter

Reduce unnecessary exposure

Pause, protect context, and get support.

A high score is a prompt for careful review—not a prediction. The practical details below are bundled into this page and work without contacting anyone.

If you may be in immediate dangerMove to a safer place, tell a trusted person, and do not confront a suspected monitor. If state actors may be the source of the threat, do not assume an official reporting channel is neutral.

SELF-GUIDED · DIGITAL SAFETY EFF Surveillance Self-Defense
Start with a security plan: identify what must be protected, who may want it, likely consequences, and which risks are most probable. Then secure the highest-risk accounts and devices first.
- Use unique passwords stored in a password manager.
- Prefer app- or security-key MFA; save recovery codes offline.
- Update devices, review active sessions, and encrypt devices and backups.
FRAMEWORK · ORGANIZATIONS Freedom House Digital Security Framework
For a newsroom, civic group, or human-rights organization: separate an active incident from longer-term hardening. Match support to the context rather than applying one universal checklist.
- Emergency: stabilize the incident and protect access to essential data.
- Short term: assess threats, train people, and adopt targeted controls.
- Long term: document roles, response steps, backups, and recurring review.
U.S. CIVIL LIBERTIES · LOCAL INTAKE ACLU state affiliates
The ACLU has staffed affiliates across the United States, Washington, D.C., and Puerto Rico. Intake rules and case selection vary, and an affiliate may refer a matter rather than represent it.
- Find the affiliate for the place where the incident happened.
- Prepare a short chronology, dates, notices, screenshots, and deadlines.
- For an urgent court or filing deadline, also seek a local lawyer immediately.
U.S. LEGAL REFERRAL · MASS DEFENSE National Lawyers Guild
The NLG referral directory lists member lawyers by location and area of work. NLG says it does not endorse listings and referrals may charge. Some volunteer chapters also provide protest support, legal observers, or local hotlines.
- Check jurisdiction, relevant practice area, fees, and conflicts before hiring.
- Ask a local chapter separately about mass-defense support.
- Keep legal deadlines moving while waiting for any referral response.
24/7 · TECHNICAL INCIDENT HELP Access Now Digital Security Helpline
Free remote digital-security assistance for eligible civil-society groups, activists, researchers, journalists, bloggers, and human-rights defenders. It covers account, device, communication, cyberattack, and surveillance-technology concerns—not legal or physical-safety cases.
- From a device you trust, describe the issue without sending unnecessary sensitive files.
- Expect an eligibility check and a move to a secure communication channel.
- The service says it is available 24/7 in ten languages.

External details can change. If you go online, verify the organization name and destination before sharing information. These resources do not replace advice from a lawyer or a personalized security assessment.

Exposure Meter is free for the people who need it most. Donations keep it that way.

Choose the support that fits.

The full scanner remains available to everyone. Paid plans fund deeper support, regional adaptation, and organization-wide rollout.

Free

The complete private safety scanner for individual use.

Full safety scanner
Unlimited scans
Published scoring methodology
Local-first always

Free forever for individuals.

Use the free scanner

Pro

$6/month

Everything in Free, plus tools for ongoing review.

Scan history and comparison over time
Regional lexicon packs
Priority support

Pro never changes the privacy architecture — your posts still never leave your device.

Choose Pro

Organization

$99/month

For newsrooms, NGOs, and legal-aid groups.

Team training materials
Guided rollout support
Custom regional lexicon pack
Invoice billing

A practical rollout path for teams working with sensitive material.

Talk to us

Readable by design.

Authoritarian movements have historically treated dissent, opposition ties, and criticism of leaders as threats. A future system could scan those signals at far greater scale. This defensive model makes that possibility visible without claiming to reproduce any real regime's methods.

What the score measures

Eight categories of machine-legible text signals: political figures, civic activity, protest or organizing language, state institutions, criticism of power, time and place, network references, and emotional intensity, plus direct identifiers. Critical language gets the highest published weight; substantive criticism and generic pejoratives are related to an already-attributed leader, officeholder, party, regime, or government reference within the same post. The two rules remain separately traceable in the explanation.

Published weights and word lists

Criticism of power · 10 points per match: corrupt, corruption, dictator, dictatorial, tyrant, tyranny, fascist, fascism, authoritarian, autocrat, resign, impeach, impeachment, criminal, liar, lying, stolen, rigged, oppressive, oppression, crackdown, censorship, censored, unconstitutional, injustice, abuse of power, police state, anti-government, dissent, dissident, oppose/opposed, criticize/criticized, criticise/criticised, condemn/condemned, denounce/denounced, resist, resistance. +20 points per post where one appears with a political figure or state institution.
Pejorative attributed to power · 10 points per match · English list v1.0: stupid, incompetent, corrupt, liar / a liar, pathetic, disgrace / a disgrace, failure / a failure, joke / a joke, criminal / a criminal, evil, tyrant / a tyrant, fraud / a fraud. This rule only fires after the existing layer attributes a public figure, party, role, or institution in the same post. It uses the same context window as criticism of power, checks both word orders, and gives any overlapping term span to one rule only. A narrow literal negation guard blocks the rule when not, isn't, never, or hardly appears between the attributed entity and pejorative in the same clause; this is not a claim of stance understanding. Identity-based or protected-characteristic slurs and profanity-only terms are excluded.
Protest & organizing · 8: protest, march, rally, demonstration, organize, organizing, organizer, strike, boycott, resistance, mobilize, walkout, picket, direct action.
Political figures · 7: president, vice president, VP, prime minister, chancellor, minister, head of state, ruling party, governing party, opposition, opposition party, party leader, political leader, Democratic Party, Labor/Labour Party, Conservative Party, Green Party, Nazi Party, National Socialist, fascist party—plus the built-in parties and starter leader names below and optional names or aliases you enter.
Built-in political parties · same 7-point weight: Republican Party/GOP/Republicans; Labour; Renaissance; Christian Democratic Union/CDU; Brothers of Italy/Fratelli d'Italia/FdI; Liberal Party/Liberals; Liberal Democratic Party/LDP.
Starter world leaders · same 7-point weight: Narendra Modi/Modi; Xi Jinping/Xi; Vladimir Putin/Putin; Donald Trump/Trump; Andy Burnham/Burnham; Keir Starmer/Starmer; Emmanuel Macron/Macron; Sébastien Lecornu/Sebastien Lecornu/Lecornu; Friedrich Merz/Merz; Giorgia Meloni/Meloni; Mark Carney/Carney; Anthony Albanese/Albanese; Javier Milei/Milei; Claudia Sheinbaum/Sheinbaum; Luiz Inácio Lula da Silva/Lula; Cyril Ramaphosa/Ramaphosa; Prabowo Subianto/Prabowo/Subianto; Recep Tayyip Erdoğan/Erdogan; Mohammed bin Salman/bin Salman; Kim Jong Un/Kim Jong-un; Mojtaba Khamenei/Mojtaba/Khamenei; Ali Khamenei; Masoud Pezeshkian/Pezeshkian; Alexander Lukashenko/Lukashenko; Delcy Rodriguez/Delcy Rodríguez/Rodriguez; Nicolás Maduro/Nicolas Maduro/Maduro; Miguel Díaz-Canel/Diaz-Canel; Daniel Ortega/Ortega; Abdel Fattah el-Sisi/el-Sisi/Sisi; Min Aung Hlaing; Ahmed al-Sharaa/al-Sharaa; Viktor Orbán/Orban; Ilham Aliyev/Aliyev; Ursula von der Leyen/von der Leyen; Lee Jae-myung; Shigeru Ishiba/Ishiba; Sanae Takaichi/Takaichi.
Time & place · 7: meet at, meeting at, join us at, tonight, tomorrow, this evening, at noon, at midnight, location, address, on my way, arriving at; detected dates, times, and street addresses.
Civic activity · 5: election, vote, voting, ballot, campaign, petition, policy, council, hearing, advocacy, activist, democracy, civil rights.
State institutions · 5: government, administration, agency, police, court, congress, senate, legislature, ministry, military, governor, mayor.
Direct identifiers · 5: Email addresses, phone numbers, URLs, and public @handles; street addresses add 9 points.
Network references · 4: join us, our group, members, organizers, coalition, chapter, bring a friend, share widely, signal group, group chat, and @mentions.
Emotional intensity · 3: angry, furious, outraged, afraid, terrified, hate, betrayed, disgusted, unacceptable, enraged, fed up; repeated punctuation and all-caps emphasis.

The built-in leader list is current as of September 2026. Former leaders are deliberately retained so historical posts still score. Leaders change; regional lexicon packs are the long-term mechanism for keeping names current.

Iran: Mojtaba Khamenei succeeded Ali Khamenei (d. February 2026) in March 2026.

Venezuela: Delcy Rodriguez has served as interim president since January 2026, when Nicolás Maduro was captured.

France has had multiple prime ministers within a year, including resignation and reappointment cycles, so its entries follow a shorter review cadence than stable entries.

Each category is capped at six weighted matches. Multiple active categories and protest-plus-place combinations add disclosed combination bonuses. Short texts are capped at 46, but criticism attributed to power—including the pejorative rule—has a 45-point floor.

Phase 2 assurance checks

Control	Evidence	Gate
Pejorative-proximity guard	11/11 corpus cases pass, including negation, unattributed-entity, historical-post, per-entity, directionality, and span-dedup checks.	Required
Pejorative list scope	English v1.0 manually reviewed: no identity-based or protected-characteristic terms; no profanity-only terms.	Required

The negation guard is deliberately narrow: it blocks not, isn't, never, or hardly only when the word appears between the attributed entity and pejorative within the same clause. It is a literal precision check, not a claim of stance understanding.

What it cannot know

This is keyword and proximity matching, not machine-learning stance analysis. It cannot reliably understand context, sarcasm, quotations, images, audience, intent, or whether a claim is true. Its negation check is intentionally limited to four literal words between a recognized power reference and a pejorative in the same clause. It assigns no ideology label, stores no profile, connects to no social platform, and compares your text with no one else's. A low score is not a promise of safety.

Safer ways to use social data

Prefer pasted public text or a platform data export. Never give a third-party tool your social-media password. If you use OAuth elsewhere, check that access is read-only and revoke it when finished.

Parsing, large exports, and safer reports

Uploads are separated into candidate posts locally, with the found and skipped counts shown before scoring. Scoring runs in short chunks so the browser can keep responding on large exports. Downloaded reports omit raw post text by default; quoting requires an explicit choice after a sensitivity warning. The local alias field is only for public officeholder, party, role, or institution names—never private people, handles, or contacts.