Why This Exists
A promising tool is not the same as a proven protection.
Survivor-facing technology can help people find information, document abuse, plan for safety, reach support, or use AI. Before an advocate recommends a product or a survivor relies on it, someone has to ask what the product actually does, what the evidence supports, and what remains unknown.
The important question is not only, “What does this tool promise?” It is, “What would someone need to know before relying on that promise?”
- Starting pointProduct promiseWhat the marketing says it does
- The missing layer
- 01QuestionsSurvivor-specific threat model
- 02EvidenceVerbatim source quotes
- 03What remains unknownGaps, follow-ups, and tests
- DestinationHuman decisionThe advocate's and survivor's choice
The decision before the decision
Useful features are only one part of the decision.
A product can meet its intended purpose and still raise questions about privacy, confidentiality, access, retention, reliability, security, or survivor safety.
The Functional Utility
A useful product feature answers one primary question: can this technology do something a survivor or program needs? It might offer private journaling, legal statute guides, emergency dialing, or safe shelter maps.
The Consequential Questions
A responsible adoption decision raises additional questions. What information does it collect? Who can access it? How long does it remain? What happens when a feature fails? What changes when the device or account may not be private?
Cross-Disciplinary Realities
These are different forms of expertise. Advocates should not have to become security engineers, privacy engineers, cloud architects, digital forensic examiners, and AI evaluators merely to know which questions deserve an answer.
What the tool is for
Intended functional utility: messaging, evidence capture, safety planning, resource navigation.
- PrivacyData collection & surveillance exposure
- ConfidentialityThird-party sharing & subpoena limits
- ReliabilityOffline behavior & failure modes
- Data AccessWho can read, export, or audit records
- SecurityKey management & credential storage
- AccessibilityAssistive tech & cognitive clarity
- AI SafetyBoundary containment & hallucination risk
- Survivor SafetyDiscretion, discovery risk & traces
Adoption & Recommendation Decision
A synthesized evaluation balancing functionality against verified safeguards and known risk profiles.
A different threat model
Domestic violence changes the assumptions technology usually makes.
A device may not be private. An account may not be controlled by one person. A change in access may be noticed. A feature that is useful in one situation can carry different consequences in another.
In conventional consumer software, developers assume a benign environment: the person holding the device owns it, nobody else has physical access or account passwords, and notifications or location services are purely convenient.
In victim services, those assumptions break down. An abusive person may have physical access to a survivor's device or remote access to activity on it. Download histories, backups, synced records, communications, location information, and other digital traces can matter when technology is being monitored. Furthermore, cutting off access or abruptly removing monitoring technology can alert an abusive person and may increase danger.
| Common product assumption | Survivor context may add |
|---|---|
The device belongs to one user | Another person may inspect or control it |
The account is private | Credentials or family accounts may be shared |
Notifications help the user | Notifications may reveal activity |
Location improves convenience | Location can become surveillance data |
Cloud sync protects information | Synced copies can create additional exposure |
Deleting something reduces risk | The deletion itself may be noticed |
A hidden screen means discreet use | Other traces may remain elsewhere |
Presumed Benign Context
Assumes single-user privacy, private accounts, and benign environment.
- No physical adversary
- Notifications are harmless
- Deletion leaves no suspicion
Reality of Coercive Control
A device may be inspected, shared, backed up, or tracked by an abusive person.
- Physical Device
- Shared Accounts
- Cloud Backups
- Family Plans
- Location Traces
- Notifications
Claim ≠ Conclusion
Words like “private” and “encrypted” are questions, not conclusions.
Marketing language may describe a real safeguard. It still does not answer every question a survivor or advocate may need answered.
“Encrypted”
Encrypted from whom? Who holds the keys?
Who can still decrypt or access the records, and what metadata remains visible?
“Anonymous”
What information is collected before, during, or after anonymization?
Are IP addresses, device identifiers, or interaction timestamps retained?
“Delete anytime”
Deleted from the screen only, or also from backups, logs, and vendors?
Does deletion propagate across server replicas, analytics providers, and exports?
“Hidden”
Hidden where? What traces remain outside the app?
What appears in app store history, battery usage, account sync, or notification logs?
“Emergency help”
What happens with no signal, backgrounded apps, or failed delivery?
How does the system fail when cellular data, GPS, or power are constrained?
“AI-powered support”
What reaches the AI model, and what happens to the conversation?
Can the AI hallucinate legal steps, and are chat transcripts logged for model training?
The research gap
Finding information is not the same as knowing what it proves.
Search engines, product websites, app stores, privacy policies, news coverage, and AI assistants can all help surface information. The harder task is determining what each source actually establishes and what it cannot.
Company Documentation
A company's own documentation tells you what the company says. It reflects intended architecture and marketing commitments, but has not been independently verified.
Privacy Policies
A privacy policy is important legal evidence, but it often describes corporate permissions rather than technical limits, and may omit crucial details about data deletion or third-party SDKs.
Generative AI Assistants
Generative AI tools can confidently draft fluent text, but risk generating incorrect conclusions or fabricated citations without grounded verification. AI tools used around survivors create accuracy and safety risks.
- Product WebsiteCompany statement
- Privacy PolicyLegal commitment
- App StoresDeveloper metadata
- Outside ReportingIndependent review
- Public RecordsRegulatory & corporate
- AI & Web SearchDiscovery, not proof
What does this source establish?
Telling observations apart from promises, marketing claims, and ungrounded output.
- Public information answers thisA reviewed source gives an answer. It may describe a safeguard or a problem.
- Company-provided information onlyOnly the company describes it. No independent confirmation was found.
- Sources disagreeReviewed sources give conflicting information.
- Not answered by the sources reviewedNo reviewed source answered it. That does not mean it does not exist.
- Needs an answer from the companyOnly the company can explain how this works.
- Needs a safe test on a spare deviceCheck it on a spare device using made-up information.
What people may not know to ask
The hardest risks are often hiding in the second question.
A product may answer the obvious question while leaving the consequential one unresolved.
- 01
Who else can see the information?
Cloud staff, analytics vendors, subprocessors, or law enforcement
- 02
Where does it go after someone enters it?
Downstream AI APIs, logs, third-party databases, and sync queues
- 03
What remains after “Delete”?
Immediate device wipe versus archived backups, logs, and account caches
- 04
What happens when the feature fails?
Offline behavior, silent SOS drops, location timeouts, and crash errors
- 05
What traces remain outside the app itself?
App store histories, push notifications, bank statements, and battery logs
- 06
What has actually been verified, and what is still only a company statement?
Independent technical review versus unverified promotional copy
And sometimes the most important finding is that no reviewed source answered the question at all.
“Not answered” is not the same as “does not exist.” A safeguard may be present but undocumented, a page may be unavailable, or the answer may require asking the company or conducting a hands-on test. Responsible review makes that uncertainty visible instead of turning it into an unwarranted conclusion.
A better review standard
Good review needs more than a checklist.
The value comes from connecting technical evidence to survivor context, making uncertainty visible, and keeping the decision with a person.
- Principle 01
Survivor Context
Interpret technology through real possibilities of shared access, monitoring, coercion, and survivor choice.
- Principle 02
Evidence
Show what information supports the conclusion with word-for-word quote verification.
- Principle 03
Provenance
Make clear who said it, where it came from, and whether it was observed or self-reported.
- Principle 04
Uncertainty
Say plainly when the public record cannot answer the question, avoiding false confidence.
- Principle 05
Consequence
Explain why a technical detail could matter for survivor safety or informed choice.
- Principle 06
Validation
Turn unanswered questions into company follow-ups or appropriate spare-device testing.
- Principle 07 · The capstone
Human Judgment
Leave the recommendation with the person who understands the survivor, the program, and the specific circumstances. Technology supports the decision; it never replaces it.
The missing layer
That gap is why we built Survivor Tech Review.
It sits between “this looks helpful” and “we are ready to recommend it.”
Survivor Tech Review was built to bring survivor-specific questions, public evidence, source provenance, explicit unknowns, company follow-ups, testing needs, and human review into one place.
It helps reviewers distinguish what is known from what is claimed, what the available evidence can support from what it cannot, and which unanswered questions may matter before a survivor relies on a product.
It does not certify technology as safe. It gives advocates and other reviewers better information for a decision that remains human.
Survivor Tech Review is decision support, not automated endorsement.
- InputPromising productMarketing promises, features, intended utility
- Decision-support layerSurvivor Tech ReviewStructured analysis of public evidence & gaps
- OutcomeInformed human decisionContext-aware choices made by advocates & survivors
Survivor Tech Review was created through work at the intersection of domestic violence systems, survivor-centered practice, technology, and software development.
The decision stays with people.
Technology can provide evidence. Software can organize it. AI can help analyze it. None of those things understands a survivor's complete circumstances.
Survivor Tech Review is designed to support the judgment of advocates, programs, and survivors, not replace it.
Evidence behind this approach
Survivor Tech Review's approach draws on published guidance and research concerning survivor privacy, technology-facilitated abuse, digital services, technology selection, and responsible AI use.
- NNEDV Safety Net · Choosing and Using Apps
Supports device access, app traces, testing safety features, location concerns, and source expertise.
- NNEDV Safety Net · Choosing a Vendor and Choosing a Platform
Supports privacy, security, accessibility, vendor practices, and platform evaluation by victim-service programs.
- NNEDV Safety Net · Best Practice Principles for Digital Services
Supports survivor privacy, minimal collection, informed choice, platform-specific risks, and planning for unexpected behavior.
- NNEDV · Artificial Intelligence & Victim Services / Survivor's Guide to AI
Supports AI-specific privacy, safety, trust, and accuracy considerations in victim advocacy.
- Federal Trade Commission (FTC) · Stalkerware: What to Know
Supports device surveillance, escalation risks, safer-device considerations, and digital evidence preservation.
- NIST AI 600-1 · Generative AI Profile
Supports risks from confident false outputs, including fabricated logic or citations in automated evaluation.
- Scoping Reviews of Technology-Facilitated Abuse (PubMed)
Supports the documented role of mobile devices, surveillance apps, spyware, cameras, and social networks in intimate partner abuse.