AI TRUST ID

Command risk

Building Now

Works here: PS command-pattern preview.

PS checks for commands that run downloaded or encoded code and explains what it finds.

For example: An AI answer gives you an install command. PS points out supported download-and-run patterns before you use it.

A warning is about the command, not proof of a scam. No warning does not mean it is safe.

Use this tag

Copy an AI answer, paste it and check. No account or install needed. Commands are checked, never run.

PS in Terminal

Prefer Terminal? Download the source, save an answer as response.txt, and run the command below on Linux or macOS with Python 3.12+. No local service or privacy checker is needed. Up to 20,000 characters. No redaction is performed.

Download source

python3 scripts/check-ps.py response.txt

Check on this device

Open aitrustid.com/#person/PS and choose Check on this device. Optional: save the public app for offline use and add it to your home screen. No answer is uploaded; the phone preview performs no redaction.

Developer local service (optional)

Developer preview: Linux, Python 3.12+ and an installed, working Podman engine. First setup downloads pinned dependencies. Read the full guide for private build-network options and the older manual Docker Compose path. This command does not install an engine; no Chrome Web Store release is offered.

Download source · Full setup guide

Get the source
Download and unzip the source, then open its folder. Or use Git:
```
git clone https://github.com/knwvmbrr/aitrust-id.git
cd aitrust-id
```
Start your own checker
On Linux with Python 3.12+ and a working Podman engine, run this from the source folder. It creates a private installation and checks that it works:
```
python3 scripts/setup-local.py --engine podman
```

Check an answer

Save an answer as response.txt, then run:

LOCAL_STATE="$HOME/.config/aitrust-id/local-stack"
"$LOCAL_STATE/checker/bin/python" scripts/aitrust.py check response.txt --env-file "$LOCAL_STATE/runtime.env"

A PS finding asks you to review a command pattern. No finding means no supported pattern was found, not “safe.” If a check cannot finish, it says so.

Optional Chrome tags: follow docs/gateway-authentication.md in the downloaded repository to set up gateway signing. Open chrome://extensions, enable Developer mode, choose Load unpacked and select the extension folder. In its settings, choose your gateway public-key JSON, confirm its fingerprint separately and enter your local token. Turn on Enable checks on ChatGPT when ready. Keep your services running. Tags appear below supported answers; click for a short explanation. Unsigned or altered results stay unavailable. Pause or reset at any time.

Your answer goes to your own local service, not this website or a training database. Keep your token private. Do not paste it into a public report.

Repeat a saved local-service check. This compares your saved findings against the same method; it does not send research data to us.

Stop the checker

python3 scripts/setup-local.py --engine podman --stop

Performance

Working preview

Development examples · not an independent accuracy study

Correct alerts

100.0% · 31/31

95% range 89.0%–100.0% · candidate lower bound 93.0%

How many flagged examples were correctly flagged.

Patterns caught

100.0% · 31/31

95% range 89.0%–100.0% · candidate lower bound 75.0%

How many labeled risk examples were found.

Correct alerts: 31
False alerts: 0
Missed patterns: 0
Correctly unflagged: 55

Test breakdown

Development input groups, not a representative accuracy study.

Encoded content: 6 examples · 0 false alerts · 0 missed patterns
Downloads saved to files: 17 examples · 0 false alerts · 0 missed patterns
Ordinary answers: 5 examples · 0 false alerts · 0 missed patterns
Other code forms: 5 examples · 0 false alerts · 0 missed patterns
Other download commands: 42 examples · 0 false alerts · 0 missed patterns
Quoted or discussed downloads: 11 examples · 0 false alerts · 0 missed patterns

Local service speed

Includes redaction and tagging. The phone preview uses a different route.

256 characters: 54 ms
2,000 characters: 107 ms
8,000 characters: 472 ms
20,000 characters: 1.70 s

20 repeats per profile on Linux (4 logical CPUs). Bars show the slowest profile’s warm p95, not a speed guarantee.

Measured 2026-10-14 · synthetic inputs

Download service timings

Efficiency tests compare warm rule checks and temporary Python allocations on the same computer, not accuracy or whole-service speed. Download paired efficiency tests

86 development examples. Range shows sampling uncertainty; tuned examples do not establish real-world accuracy. Independent release validation: pending.

Measured 2026-10-13

Next test: A new frozen, independently labeled set, followed by release acceptance.

Job

Find supported network-to-shell and encoded-payload execution patterns in an identified response.

Outcome

Read the matched pattern and its limits before acting.

Claim

A supported command-risk pattern was found. The adopted PS name previously overclaimed scam detection; this preview uses a bounded explanation.

Method

Local lexical matching. Pipe and substitution methods check stdout routing and the execution layer; supported comments, immediate warnings and literal display forms are excluded. No payload is fetched or executed. The evaluator source hash identifies the exact method.

Input

Text, currently through the Chrome/ChatGPT development adapter and authenticated local gateway.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

curl or wget piped to sh or bash, with optional sudo
Bounded curl/wget stdout command, process and backtick substitution into supported execution commands
eval/exec receiving base64.b64decode, atob, or bytes.fromhex
Exact spans in normalized redactor output

Limits

Does not identify a scam, malicious intent, or human/AI authorship
Does not execute commands or parse every shell/language form
Quoted or escaped command names and decoder aliases can be missed; eight syntax/token-preserving spelling tests bypassed the current matcher
No finding is not a safety clearance; one live Homebrew response verified, broader live-site coverage remains open

Validation

Current context-v6 development regression: 86 cases across five active sets, TP 31 / FP 0 / FN 0 / TN 55. These were used during development, not an independent holdout. Development Wilson 95% lower bounds are 0.890 for precision and recall; the candidate precision gate fails. Independent labels and calibration are missing. Current context-v6 service staging passed 32 real HTTP checks, including redaction binding, dependency failures and recovery. Eleven additional real HTTP fault checks cover saturation, disconnect and timeout. These are engineering checks; broader live adapter, human usability and independent accuracy gates remain open.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state
Local redaction before evaluation
Independent use/mention labels

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Candidate lower bounds 0.93 precision / 0.75 recall need accepted sampling and independent labels. Browser, accessibility, security, challenge/correction and latency gates remain open.

T-PS · Not release validated

PII

Personal information redacted

Building Now

Local setup required. PII does not run in this website checker.

Keep detected personal details out of the local checker.

For example: An answer includes an email address. The local service can remove detected values before checking the rest of the text.

Some sensitive details can be missed. This tag does not mean all private information is gone.

Use this tag

Check a draft with your own local redactor before you share it. Review the result and make any edits yourself. This is a local-service preview, separate from the website’s on-device PS checker.

Download source · Full setup guide

Get the source
Download and unzip the source, then open its folder. Or use Git:
```
git clone https://github.com/knwvmbrr/aitrust-id.git
cd aitrust-id
```
Start your own checker
On Linux with Python 3.12+ and a working Podman engine, run this from the source folder. It creates a private installation and checks that it works:
```
python3 scripts/setup-local.py --engine podman
```
Review a draft before sharing
Save text as draft.txt. This optional privacy check uses your local redactor and does not depend on PS. It does not edit or send the draft:
```
LOCAL_STATE="$HOME/.config/aitrust-id/local-stack"
"$LOCAL_STATE/checker/bin/python" scripts/aitrust.py privacy draft.txt --env-file "$LOCAL_STATE/runtime.env"
```

A finding means supported detections were masked during this local check. Your original draft is unchanged. No finding does not mean all private details are absent.

Your answer goes to your own local service, not this website or a training database. Keep your token private. Do not paste it into a public report.

Repeat a saved local-service check. This compares your saved findings against the same method; it does not send research data to us.

Stop the checker

python3 scripts/setup-local.py --engine podman --stop

Performance

Working preview

English redaction development checks · not independent validation

Redaction test breakdown

Synthetic examples on the English service, not the phone preview.

34/34 expected identifiers fully masked in English test examples. 2/38 examples had extra masking.

Dates and times: 3/3 exact finds · 0 missed · 2 extra candidates
Email addresses: 13/13 exact finds · 0 missed · 0 extra candidates
IP addresses: 3/3 exact finds · 0 missed · 0 extra candidates
Places: 3/3 exact finds · 0 missed · 0 extra candidates
Names: 5/5 exact finds · 0 missed · 0 extra candidates
Phone numbers: 5/5 exact finds · 0 missed · 0 extra candidates
Web addresses: 2/2 exact finds · 0 missed · 14 extra candidates

1 extra candidates from other recognizers.

Candidates can overlap. An extra candidate is not necessarily an extra redaction. Extra masking can remove ordinary text.

Four out-of-coverage challenges: 2/4 expected identifiers fully masked; 1 partially masked and 1 untouched.

Compact IP label checks: 6/6 expected addresses masked in six additional synthetic examples; 3/6 examples had extra masking (12 characters).

Download compact IP tests

38 developer-authored English examples and four out-of-coverage challenges. Exact entity/span matches and actual masking are different measures; no population or independent accuracy claim. Other identifiers or sensitive context may be missed. Review before sharing.

Download redaction observations

Local service speed

Includes redaction and tagging. The phone preview uses a different route.

256 characters: 54 ms
2,000 characters: 107 ms
8,000 characters: 472 ms
20,000 characters: 1.70 s

20 repeats per profile on Linux (4 logical CPUs). Bars show the slowest profile’s warm p95, not a speed guarantee.

Measured 2026-10-14 · synthetic inputs

Download service timings

Independent release validation: pending.

Measured 2026-10-11

Next test: Independently review recognition, extra masking and real-user operation.

Job

Redact detected personal information before local evaluation or a selected draft review.

Outcome

The record states which entity types were detected. Privacy-only review does not call PS.

Claim

Detected entity values were redacted in the local check. Some categories, such as URLs, can contain public information. This does not certify that all sensitive information is gone or change your original draft.

Method

Local Presidio recognition with a build-provisioned English model and offline email-domain validation. Privacy mode selects the redactor independently.

Input

Text you explicitly provide to your local gateway.

Output

Entity type/count metadata without private values. The complete pipeline forwards redactor output to PS; privacy-only review does not.

Supported scope

Email recognition exercised against the real service
Detected redaction precedes evaluation
Manual privacy-only draft checking
Offline suffix data; no runtime model download

Limits

Personal or sensitive information can remain undetected
English assets are the current development target
This manual check does not intercept or prevent sending
Neither original nor redacted content should be submitted to a public report

Validation

Real email redaction, pipeline ordering, source hashes, and synthetic log markers checked. The independent privacy route requires its current service evidence; independent recognition accuracy and human acceptance remain open.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A required redaction failure withholds the result. An evaluator outage does not block an explicitly selected privacy-only check.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-PII · Not release validated

Non-Fiction

Coming Soon

Planned · No checker available yet.

See how an answer lines up with the references you choose.

For example: Compare an answer with your handbook and see the passages that support its claims.

Planned. A match is only as reliable as the reference; it is not proof of truth.

What you can do

Suggest a claim and the reference passage that supports it.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Benchmark supported and unsupported claims against a declared reference corpus.

Job

Check whether claims agree with supplied reference material.

Outcome

See which reference passages support a claim and where support is missing.

Claim

Consistent with an attached reference corpus; not a certificate of universal truth.

Method

Proposed local retrieval and claim-to-passage matching.

Input

Content and a user-supplied local reference corpus.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Planned local corpus support and cited evidence

Limits

Reference material may be incomplete, wrong, or outdated
No attached corpus means no supported corpus verdict
Consistency is not proof of factual truth

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state
User-supplied reference corpus

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-NF · Not release validated

Fiction / Fabricated

Coming Soon

Planned · No checker available yet.

See where an answer disagrees with your references.

For example: Compare a claim with your handbook and read the passage that contradicts it.

Planned. A disagreement does not prove dishonesty, and the reference could be wrong.

What you can do

Share a claim and a reference passage that disagree.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Test actual reference contradictions separately from missing evidence.

Job

Identify claims contradicted by supplied reference material.

Outcome

See the contradictory passage and the scope of the disagreement.

Claim

Reference contradiction or an unverified citation requires bounded wording; absence of support alone does not establish fabrication.

Method

Proposed local corpus contradiction and citation checks.

Input

Content and a local reference corpus.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Planned corpus contradiction with inspectable spans

Limits

An incomplete corpus cannot prove a citation does not exist
A contradiction can reflect an incorrect reference
No intent or dishonesty finding

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-FI · Not release validated

Hallucination possible

Coming Soon

Planned · No checker available yet.

Take a closer look at claims that may not hold up.

For example: Review a specific claim alongside its supporting evidence, rather than judging how confident the answer sounds.

Planned. Confident wording, dates or repetition alone cannot show that a claim was invented.

What you can do

Suggest a claim that needs checking, with a source that explains why.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Validate specific indicators against independently labeled unsupported claims.

Job

Evaluate validated indicators of unsupported or inconsistent claims.

Outcome

See the questionable claim, its indicators, and uncertainty.

Claim

Possible unsupported content, only if separately validated indicators justify it.

Method

Proposed contradiction, unsourced specificity, repetition, and hedge analysis. None ships in the current evaluator.

Input

Content, with references where the method requires them.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Research scope only

Limits

Dates, precise writing, repetition, or missing hedges do not establish hallucination
Cannot be emitted by the current runtime
Independent labels are missing

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-HP · Not release validated

Pressure cues

Coming Soon

Planned · No checker available yet.

A little context when a message asks you to act quickly.

For example: A message says “act now.” MT would help you look at the wording and what is happening around it.

Urgent messages can be helpful. This planned tag would highlight wording, not judge the person behind it.

What you can do

Share an example of urgent wording and the context that makes it helpful or pressuring.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Compare helpful and pressuring messages, with independent reviewers checking the context.

Formal tag name

Manipulative tactic

Job

Identify independently validated framing patterns that may pressure a reader.

Outcome

Read the specific wording and its possible effect without an intent accusation.

Claim

An observed framing pattern, not proof of manipulation or motive.

Method

Proposed urgency, dilemma, and authority-framing analysis.

Input

Content and its relevant context.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Research scope only

Limits

Urgency and persuasion can be legitimate
Context and disagreement must be reported
The MT-specific human-agreement gate must not block other tags

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-MT · Not release validated

Independently verified

Coming Soon

Planned · No checker available yet.

See what an independent person actually checked.

For example: Read a reviewer’s evidence, what they reviewed and what their review did not cover.

Planned. A signature records a review; it does not guarantee every claim is true.

What you can do

Help define what a reviewer should record, and how to correct or withdraw a review.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Verify reviewer identity, independence, artifact binding and challenges.

Job

Record an identified reviewer’s independent check and attestation.

Outcome

Know what was checked, by whom, with what scope and evidence.

Claim

A person attested to a defined review; the signature alone does not establish truth.

Method

Human attestation only, with a defined signing and challenge policy.

Input

An identified artifact, reviewer evidence, and authorized attestation.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Human review is the intended basis

Limits

Never machine-assigned
Reviewer independence and review scope must be checked
Not a universal guarantee

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-IV · Not release validated

Full AI augmentation

Coming Soon

Planned · No checker available yet.

See evidence that AI generated the content.

For example: Inspect a recorded generation event or authenticated content credential tied to the item.

Planned. Writing style or missing human evidence cannot prove AI origin.

What you can do

Suggest a generation record that could show where an item came from.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Test positive generation evidence, binding and unknown-source behavior.

Job

Report positively supported generation provenance.

Outcome

Inspect the provenance event or authenticated manifest behind the claim.

Claim

Observed provenance supports an AI-generation assertion within the witnessed scope.

Method

Proposed capture-point provenance and authenticated content credentials.

Input

Positive provenance evidence and an identified artifact.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Planned authenticated provenance evidence

Limits

Missing human evidence must never imply AI origin
Typing patterns, disability, language style, or repetition are not origin proof
A trusted signature is evidence of its signer’s statement, not independent truth

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-FA · Not release validated

Partially augmented

Coming Soon

Planned · No checker available yet.

See recorded AI contributions and later edits.

For example: Follow a consented record of an AI draft and the editing that happened afterward.

Planned. It describes observed contributions, not a person’s identity or integrity.

What you can do

Suggest a consented example of an AI draft and the edits made afterward.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Test contribution/edit receipts, consent and assistive-input handling.

Job

Report supported AI contribution and subsequent observed editing.

Outcome

Understand which contribution and editing activity the evidence actually establishes.

Claim

Observed augmentation within a defined composition window; not a biometric identity or integrity score.

Method

Proposed consented provenance and edit-event receipts.

Input

Consented composition evidence and artifact binding.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Planned positive contribution evidence and accessible input handling

Limits

No negative verdict from absent human evidence
Timing must be suppressed for assistive input; ordinary templates are not automation proof
Raw event privacy and signer integrity need independent review

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-PA · Not release validated

UNK

Unknown

Coming Soon

Planned · No checker available yet.

Know when there is not enough evidence to decide.

For example: A check finishes, but the available evidence does not support a conclusion.

Planned as a tag. Uncertainty is different from a check that failed or could not run.

What you can do

Share a case where a checker should say it does not have enough evidence.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Test insufficient evidence separately from unavailable and unsupported checks.

Job

Represent insufficient evidence without turning it into a negative verdict.

Outcome

Distinguish uncertainty from a check that failed or was unsupported.

Claim

Insufficient evidence for an asserted tag; no origin or safety conclusion.

Method

The adopted UNK code is overloaded. A six-state result contract is proposed; local preview UI already distinguishes these states.

Input

A valid evaluation result and its abstentions.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Current preview distinguishes pending, finding, no finding, uncertain, unsupported, unavailable

Limits

UNK itself is not emitted by the current runtime
An outage must never masquerade as evaluated uncertainty
No finding is not a clearance

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-UNK · Not release validated

PII+

Personal information outbound

Coming Soon

Planned · No checker available yet.

Catch detected personal details before you send a message.

For example: Notice an email address in a draft, then choose whether to edit it or send it.

Planned. Detection can miss details; the choice to edit or send stays yours.

What you can do

Try the manual draft privacy check in the PII instructions. A browser reminder or send interception is still planned.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Test pre-send detection, explicit choice and fail-safe interception.

Job

Warn before detected personal information is sent to an AI service.

Outcome

Review detected sensitive content and decide whether to redact or continue.

Claim

A detected outbound privacy risk, with user control.

Method

Planned local input recognition and consented pre-send intervention.

Input

User input before submission to a supported service.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Preserved scope; no outbound interception runs today

Limits

Detection can miss information
Must not submit or change content without a clear user action
Site forms are not an implementation of this tag

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-OUT · Not release validated

Unsafe command

Coming Soon

Planned · No checker available yet.

A proposed clearer name for the command warnings in PS.

For example: The same bounded command-risk observation could have its own code, without implying a scam.

Proposal only. UC is not issued; existing PS records keep their meaning.

What you can do

Review the proposed name and command scope in How it works.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Agree the proposed code contract and independently validate command contexts.

Job

Create an accurate successor code for supported command-risk observations.

Outcome

Understand the observed command risk without a scam implication.

Claim

Proposed successor to the bounded command portion of PS.

Method

Proposed in RFC-0004; current development still emits PS.

Input

Supported text/code instructions.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Proposal and existing PS development methods

Limits

Not adopted and not emitted
No rename can silently change existing records
Requires independent context validation

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-UC · Not release validated

Scam pattern

Coming Soon

Planned · No checker available yet.

Explore warnings about possible scam patterns.

For example: Review a message asking for an urgent payment, with each possible indicator explained in context.

Research proposal. Ordinary requests can look similar; no scam detector is available yet.

What you can do

Suggest a suspicious message and a similar everyday request to compare.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Measure fraud-indicator false alarms, misses and legitimate look-alikes.

Job

Research validated co-occurring fraud indicators in context.

Outcome

See possible fraud indicators and what they do or do not establish.

Claim

A proposed possible scam pattern, never an unqualified accusation.

Method

Pressure, value-transfer, identity, and channel-switch indicator families. Co-occurrence is an unmeasured hypothesis, not independent corroboration.

Input

Content and contextual evidence.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Research proposal only

Limits

Weak indicators may share one cause
News, quotations, and legitimate requests can contain the same terms
Recall tradeoffs must publish false positives and allow dismissal

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-SC · Not release validated

Automation pattern

Coming Soon

Planned · No checker available yet.

Explore evidence that a workflow was automated.

For example: Review positive, consented observations instead of guessing from someone’s writing style.

Research proposal. Repetition, timing, templates or missing human activity alone are not proof of a bot.

What you can do

Suggest verifiable automation evidence and a human example that could look similar.

Performance

Not running

No measured results for this tag yet.

Independent release validation: pending.

Next test: Test positive automation evidence against human and assistive-input counterexamples.

Job

Research positive evidence of automation without inferring it from absent human activity.

Outcome

See validated observations without calling ordinary human repetition a bot.

Claim

Proposed automation observation; cannot establish automation from style alone.

Method

Candidate timing and duplicate-construction indicators require validation; no detector exists.

Input

Positive, consented evidence within a defined observation scope.

Output

A versioned tag record with subject binding, method, evidence, limitations, and a result state.

Supported scope

Research proposal only

Limits

Templates, repetition, timing, and learned writing patterns are not automation proof
Absent human evidence never becomes a bot verdict
Does not ship if indicators cannot separate ordinary human behavior

Validation

Not independently release validated.

Cost

Free personal workflow. No paid detector, evidence, or verification gate.

Privacy

This catalogue does not evaluate your content. The personal reference implementation uses local services; detected redaction can miss sensitive information.

Dependencies

Versioned claim and evidence contract
Independent review for this tag
Accessible presentation and a clear failure state

Failure

A failed or unsupported check must not become a finding. A problem with this tag must not disable unrelated tags.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release.

Release requirements

Publish agreed acceptance criteria, run independent validation, and verify the complete user workflow before release.

T-BT · Not release validated

Enterprise

FLEET

Fleet tags

Coming Soon

Planned offering. No team service is running.

Bring tag tools to your team’s devices.

For example: An IT team rolls out the same checker without changing what its tags mean.

Planned. Managing devices must not weaken a tag’s evidence standard.

What you can do

Review the extension permissions in the public source, then tell us how your team manages installs and updates.

You can review the extension’s access to chatgpt.com and your local gateway (127.0.0.1:8787) in the public source. Team rollout is still planned.

Performance

Not running

Organization capability; performance not measured.

Independent release validation: pending.

Next test: Test deployment, version rollback and per-tag policy isolation.

Job

Deploy and manage tag tools across an organization.

Outcome

Apply deployment policy without changing what a tag is allowed to claim.

Claim

A deployment and policy capability. It changes who runs a tag and where, never what a tag asserts or how confident it is.

Method

Planned managed configuration layered over the existing extension and local checker. No server-side evaluation, and no policy hook that can alter a tag threshold or claim.

Input

Organization deployment policy and the existing extension/local-checker configuration.

Output

A deployment state per device and an unchanged tag record contract.

Supported scope

Available now — inspect storage permission, loopback gateway access and ChatGPT content-script access in the manifest
Available now — try the personal on-device PS checker without an account
Planned — managed installation and updates with a stable release identity
Planned — choose which tags run for each group

Limits

The development extension is side-loaded; there is no tested managed install or update route yet
Team settings must not silently redefine a tag claim, threshold or evidence standard
Rolling out a tool does not independently validate its findings
Underlying personal detection and evidence remain free

Validation

No managed deployment has been built or tested. The permission surface claim above is verifiable from extension/manifest.json in the public repository.

Cost

Proposed paid organization offering. Pricing is not set, there is no checkout, and no subscription or contract is available. The personal detection, evidence and verification capabilities stay free and open.

Privacy

The intended rollout must preserve F-086: no extension or service telemetry. Today the content script reads supported ChatGPT answers and sends them to your local gateway. A managed configuration must not quietly add a reporting channel.

Dependencies

Tested managed distribution and stable release identity
Accepted privacy terms (WP-PRIVACY)
Accepted release process (WP-RELEASE)

Failure

A failed or unsupported organization capability must not become a finding, and must not disable the personal workflow or any unrelated tag.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release and any organization commitment.

Release requirements

Define and test a managed distribution and update route with a stable release identity. Verify F-086 under managed settings and F-096 when the gateway is down, the token is invalid or a page is unsupported.

F-159 · Not release validated

public-scope.json

HOST

Host capture tags

Coming Soon

Planned offering. No team service is running.

Connect tags to more than a browser answer.

For example: With permission, a future agent could bind a tag record to a supported file on a managed device.

Planned. No host agent exists; capture needs clear consent and limited access.

What you can do

Help define which files an agent may read, how it asks permission, and how you pause or remove it.

There is no host agent to install. You can review its planned capture boundaries in F-087 and the consent, pause, reset and deletion requirements in N-004 in the public scope.

Performance

Not running

Organization capability; performance not measured.

Independent release validation: pending.

Next test: Test consented capture, permissions and artifact binding.

Job

Capture supported artifacts at a host or managed-device layer.

Outcome

Bind tag records to artifacts beyond one browser adapter.

Claim

A capture-scope capability. Widening what can be observed never widens what a tag may assert about it.

Method

Planned host agent with an explicit path allowlist and a per-session consent gate. Not built.

Input

Consented, enumerated artifact paths on a managed device.

Output

A tag record bound to a host artifact, with the capture scope recorded in the record itself.

Supported scope

Available now — the intended capture boundary is written down and reviewable before anyone pilots anything
Planned — enumerated path allowlist with no directory traversal
Planned — per-session consent with a visible indicator while capture is active
Planned — capture scope recorded inside every record the agent produces

Limits

No host agent or tested consent flow exists yet
Requires consent, least privilege and an independently reviewed capture boundary
Access must be limited to explicitly supported artifacts; wider access adds privacy exposure
Underlying personal detection and evidence remain free

Validation

No host implementation has been tested. Capture boundaries, consent and security review are required before a pilot.

Cost

Privacy

An agent may access only the agreed artifacts. N-004 covers consent, pause, reset and deletion; F-087 requires clear capture boundaries. No host agent is collecting files today.

Dependencies

Reviewed capture boundary (WP-CAPTURE)
Accepted privacy terms (WP-PRIVACY)
Accepted release process (WP-RELEASE)

Failure

A failed or unsupported organization capability must not become a finding, and must not disable the personal workflow or any unrelated tag.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release and any organization commitment.

Release requirements

Publish and independently review the capture boundary. Test consent, pause, reset, deletion, allowed paths and denied access before a pilot. The browser adapter is a separate capture route, not proof that a host agent is safe.

F-160 · Not release validated

public-scope.json

AUDIT

Audit tags

Coming Soon

Planned offering. No team service is running.

Keep a clear history of findings and corrections.

For example: A team follows which method produced a finding and how a later review changed it.

Planned. Keeping a record does not make its finding more certain.

What you can do

Open the self-hosted audit guide. Choose your retention policy, preview one detailed result, then store it with a writer account. Personal checking needs none of this.

You can run the open-source audit tool on your own Linux server. Preview each metadata record before storing it. No company-hosted audit service or automatic upload is running. “Download details” in this panel saves the offering description, not a checked answer.

Self-hosted audit guide · Try personal PS exports

Performance

Self-hosted software

Self-hosted workflow checked; hosted-service performance not measured.

Independent release validation: pending.

Next test: Independent security review and managed-service operating acceptance.

Job

Retain versioned tag records for organization audit.

Outcome

Trace record versions, review and corrections under a defined retention policy.

Claim

Keep a clear record of your team’s checks. Storing a result does not make its finding more certain.

Method

Organization-operated PostgreSQL software with separate write, read, review and deletion permissions. Appends preserve history; corrections link to earlier records.

Input

Detailed supported tag exports you choose, plus an explicitly selected metadata and retention policy.

Output

Private metadata history, fixed-choice reviews, linked corrections and verifiable exports with a separate checkpoint.

Supported scope

Available now — preview approved metadata without storing it
Available now — retain, review, correct and export records in your own PostgreSQL database
Available now — tenant and role restrictions, explicit retention, removal and encrypted-backup recovery
Personal PS summaries and optional detailed exports stay free

Limits

Requires a dedicated PostgreSQL 18 setup and an accountable operator
Hashes and evidence positions can link to sensitive information; metadata is not anonymous
A database administrator remains trusted; a local chain is not an independent public witness
No hosted retention service, organization subscription or directory integration is operating
Underlying personal detection and evidence remain free

Validation

31 actual Linux controls exercise the complete audit workflow, encrypted recovery, access refusal, expiry and personal checking while the store is stopped. These engineering checks do not independently validate a tag’s accuracy.

Cost

Privacy

No answer text, arbitrary notes, credentials or attachments enter this metadata profile. Your operator selects retention and updates recovery controls. Deletion is logical: older backups, WAL and exported copies can retain information.

Dependencies

Organization-operated PostgreSQL and existing credentials
Explicit metadata, retention and recovery policy
Versioned result exports

Failure

A failed or unsupported organization capability must not become a finding, and must not disable the personal workflow or any unrelated tag.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release and any organization commitment.

Release requirements

The self-hosted software and operator guide are available. A managed service still needs accepted operating ownership, recovery authority, monitoring and service terms. No detector release or certification follows from audit storage.

F-161 · Not release validated

SSO

Identity-managed tags

Coming Soon

Planned offering. No team service is running.

Give team members the access they need.

For example: Map directory groups to separate permissions for configuration, viewing, review and export.

Planned. Signing in does not validate a finding or make author receipts compulsory.

What you can do

Compare the proposed configure, view, review and export permissions with your team’s groups. Tell us what needs to change.

Review the proposed configure, view, review and export permissions against your team’s groups. These are design choices to discuss, not enforced roles or a working sign-in service.

Performance

Not running

Organization capability; performance not measured.

Independent release validation: pending.

Next test: Test access roles, revocation and organization separation.

Job

Manage organization access to tag workflows using directory identity.

Outcome

Assign least-privilege roles to organization workflows.

Claim

An access-control capability. Knowing who is signed in says nothing whatsoever about whether a content claim is true.

Method

Proposed directory integration with separate configuration, viewing, review and export permissions. No identity provider is connected today.

Input

Organization directory groups and the published role definitions.

Output

An authorization decision per workflow, and an access event that is kept separate from any tag record.

Supported scope

Available now — review the proposed permission split in this catalogue
Planned — separate policy administration from independent review authority
Planned — SAML or OIDC group mapping
Planned — access events kept separate from tag findings

Limits

Signing in does not prove a finding is correct
No identity provider or role enforcement has been tested
An administrator’s permissions must not silently grant independent review status
Underlying personal detection and evidence remain free

Validation

No identity provider has been integrated and no access control has been tested. The role split is a design proposal open to review.

Cost

Privacy

Directory identity is personal data. Define access-event fields and retention separately from tag findings. Response and revision binding (F-088) must not turn into an unsupported claim about a person’s identity.

Dependencies

Operations ownership (WP-OPERATIONS)
Accepted privacy terms (WP-PRIVACY)

Failure

A failed or unsupported organization capability must not become a finding, and must not disable the personal workflow or any unrelated tag.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release and any organization commitment.

Release requirements

Test least-privilege role mapping, denied access and review independence. Verify that authors can decline receipt issuance (F-128), and review key custody (F-129) and group authorship (F-130) under team identity. Login must not be used to coerce attestation.

F-162 · Not release validated

public-scope.json

SUPPORT

Supported team tags

Coming Soon

Planned offering. No team service is running.

Get a clear owner for team help and maintenance.

For example: A team knows where to ask for deployment help and who owns the agreed follow-up.

Planned. Paid support and response-time commitments are not available yet.

What you can do

Use Report an issue for a defect. For a vulnerability, choose the private security channel; do not post sensitive details publicly.

You can report defects for free today. Use the private vulnerability channel for security issues, and public issue reports for non-sensitive defects. Accessibility problems are treated as defects under the published policy; there is no paid support or promised response time yet.

Performance

Not running

Organization capability; performance not measured.

Independent release validation: pending.

Next test: Validate the actual support workflow before promising response times.

Job

Provide contracted deployment and maintenance support.

Outcome

Know who owns response and maintenance commitments.

Claim

An ownership and response capability. It is a commitment about people and timelines, not about accuracy.

Method

Planned contracted support over the existing public channels. No contract, retainer or paid tier exists.

Input

An accepted support agreement and a named organization contact.

Output

A named owner per issue class and an agreed response commitment.

Supported scope

Available now — private security disclosure through the repository
Available now — accessibility defect reporting under the published policy
Planned — a named owner for each support responsibility
Planned — agreed response commitments under a support contract

Limits

No support contract or response-time promise is offered today
Support cannot substitute for independent tag validation
Staffing and escalation need to be verified before promising coverage
Underlying personal detection and evidence remain free

Validation

No support contract has been offered, priced or honoured. The disclosure channels above are live and documented.

Cost

Privacy

An issue report can contain the content someone was checking, which is why neither original nor redacted content should be attached to a public report. Support correspondence is not a content store and must not become one.

Dependencies

Operations ownership (WP-OPERATIONS)
Published site and contact surface (WP-SITE)

Failure

A failed or unsupported organization capability must not become a finding, and must not disable the personal workflow or any unrelated tag.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release and any organization commitment.

Release requirements

Name support owners and escalation routes, verify coverage, and test the commitments before offering a contract. Check failure handling (F-096) so a service problem cannot turn into a misleading finding.

F-163 · Not release validated

public-scope.json

PRIVATE

Proprietary tag package

Coming Soon

Planned offering. No team service is running.

Add your team’s policies without changing the public meaning of a tag.

For example: A team could require an extra review while keeping the tag’s evidence and limits visible.

Planned. Private rules must not weaken or misrepresent the public standard.

What you can do

Open the local practice guide in How it works. Read its invented policy before checking the example; share requirements without confidential documents.

Try the separate term-check practice guide on a Mac or Linux computer. Nothing is uploaded and no account is required. The tool refuses public-code reuse, hidden limits, required receipts and public-threshold settings. The organization service is still planned.

Performance

Not running

Organization capability; performance not measured.

Independent release validation: pending.

Next test: Test private-policy boundaries without weakening public tag standards.

Job

Define organization-specific tag workflows and policy extensions.

Outcome

Apply a private policy with clear provenance and claim boundaries.

Claim

A private extension capability, bounded so that a private policy can add obligations but never weaken a public claim.

Method

A separate offline development tool finds exact terms in text you choose. Its private policy cannot change public tag methods. Organization rollout is planned.

Input

Your reviewed policy, a private text file and a selection that binds both files. The practice example is invented.

Output

Optional private details identify the selected namespace, policy revision and matching locations. Namespace ownership is unverified; a match is not a policy-violation verdict.

Supported scope

Available now — optional local practice with a reviewed literal-term policy
Available now — exact policy/text bindings and qualified private observations
Available now — refused public-code and privacy-weakening controls
Planned — organization ownership, each custom workflow and rollout acceptance

Limits

Display names here are offering names, not adopted standard codes
A private policy must not dilute or misrepresent the public standard, and must never reuse a public tag code for a different claim
A private extension is not a private fork of the evidence standard; thresholds move one way only, upward
Underlying personal detection and evidence remain free

Validation

The development component passed 96 focused tests and 24 actual command controls on Mac and isolated Linux. These checks cover the selected literal workflow; they do not establish namespace ownership, employee consent or independent real-world accuracy.

Cost

Privacy

A private policy must preserve consent and the public privacy commitments. Team payment does not waive no-telemetry (F-086) or capture boundaries (F-087) for the people whose content is checked.

Dependencies

Accepted privacy terms (WP-PRIVACY)
Record version contract (WP-PROTOCOL)
Independent validation (WP-VALIDATION)

Failure

A failed or unsupported organization capability must not become a finding, and must not disable the personal workflow or any unrelated tag.

Owner

Maintainers own implementation; independent reviewers own validation; the product owner accepts release and any organization commitment.

Release requirements

Complete organization ownership, agreed custom workflows, permission/export acceptance and the full conformance review before commercial rollout. The local literal prototype tests the four prohibitions; it does not validate every future private policy job.

N-010 · Not release validated