Privacy notice
Effective 2026-11-21.
Who we are
Fifteen-Twenty-Seven LLC operates OddFlag AI. This notice explains information handling for the web checker and support, and for the iOS and Android apps, including authorized testing builds when provided and public releases when offered. The free beta is for adults 18 and over in the United States. Your eligibility choice is a declaration, not independent age or location verification. A US audience does not mean information is processed only in the US. Contact support@oddflag.ai with privacy questions. The web and native screenshot paths differ as explained below. Browser extensions are outside this notice and are not offered as released services covered here.
What you submit
We process the text, URLs, selected phone number, number region and optional context you submit for a check. Context can include the organization a sender claims to represent, whether you expected contact, what was requested and whether you already acted. Online combined review sends those selected fields to our service, even if AI is off. In the web checker, separately chosen screenshot AI review also processes the selected image and interpreted text. Native apps read selected screenshot text on device; only reviewed text and separately chosen fields are used for optional online review, not screenshot pixels. Submit only material you are entitled to process. Remove passwords, security codes, full card numbers, government identification and unnecessary personal details. A selected phone number is the intended subject of a phone check; avoid unrelated contact details, private URL parameters and sensitive screenshot content.
We use the submission to show observations, uncertainty, source evidence and practical next steps. Web pattern review examines text and URL structure on our service; native pattern and structural checks run on your device. Pattern review alone does not visit websites, authenticate a sender or query an external reputation database. Separately chosen online evidence checks are described below. Rules can miss scams and warn about legitimate content; an AI interpretation or an ordinary-looking number or domain does not establish legitimacy.
No saved check history
The web application does not write submitted messages, screenshots or results to a content database or application logs. They remain in your open page for you to read. Native apps do not implement a saved review history; local drafts and temporary image handling are described below. This does not mean no records exist elsewhere: technical quota records, hosting logs, optional AI processing, reports, support correspondence and Android SDK technical data are separate.
Your separate AI choice
Optional AI review sends the content and context you choose, including a selected phone number and the evidence available to that review, to OpenAI only after your separate choice. Sources remain identified separately from AI interpretations. Web screenshot review uses OpenAI; it is not on-device image processing. AI-transcribed image links are not automatically sent for external checks: review and edit the extracted text before making a fresh lookup choice. Native screenshot text reading is on device, and optional native AI sends reviewed text and chosen context, not image pixels. You can decline AI and use pattern/structural checks and separately available online evidence. AI interpretation and transcription may be wrong.
We request no stored Response history through the API using store:false, and do not request background processing. This is not zero retention. OpenAI may retain submitted content in abuse-monitoring logs, normally for up to 30 days, with legal/safety exceptions. Temporary processing/cache and image-safety retention may also apply. See OpenAI’s data controls.
At our last account review, opt-in sharing for feedback, evaluation/fine-tuning and API inputs/outputs was disabled. Account call logging and the request’s storage choice are different controls. Disabled opt-in sharing does not mean all provider records disappear. Do not submit information because you expect immediate deletion. We do not automatically retain checks as a company training dataset.
Optional link and phone evidence
Online evidence is a separate per-check choice from AI. For eligible links you choose to check, our server sends the hostname to Cloudflare’s DNS-over-HTTPS resolver and the relevant domain to an authoritative registration-data service selected from IANA’s RDAP registry. These services receive the queried domain, our server’s connection and ordinary request metadata. We do not forward your network address or the surrounding message as a lookup field. This is not anonymity: queried names may themselves identify a person or activity, and providers keep their own records. Read Cloudflare’s resolver privacy disclosure. Registration services have their own terms and retention; we have not established a uniform retention period for all registries.
When Google Web Risk is enabled and available, the separately chosen threat check sends the eligible exact URL shown in the result to Google. Google may process the URL and connection metadata under its service terms and privacy practices. We do not promise immediate deletion or zero retention. Credential-bearing, private or sensitive-looking URLs may be blocked from threat lookup; the result identifies checks not performed. A hostname-only check is not represented as checking the original full URL. Google may receive information embedded in an eligible URL, so review the address and remove secrets before choosing a check. See Web Risk and Google privacy information.
The checker does not visit the suspicious destination, follow its redirects, run its scripts, dial a phone number or contact a sender automatically. DNS resolution, registration dates and threat-list results are distinct facts. An older registration, a DNS answer, a normal number format, a published-number match or no threat-list match does not authenticate a person or establish safety. Missing, unsupported, blocked, stale and unavailable checks are labelled.
A selected phone number is normalized and compared on our server with number-plan metadata, a bounded public FTC unwanted-call complaint snapshot and a small sourced official-contact directory. We do not send each checked number to the FTC or add your query to the public reports. The bundled public index retains reported caller IDs and aggregate counts/dates from the published data; it excludes complainant location fields. Your entered number and lookup history are not added to that index or a saved check history. A limited freshness probe may fetch the FTC publication listing, without the checked number. Snapshot updates are operator-prepared releases; the result shows its available date window, missing coverage and stale status rather than promising real-time coverage.
FTC records are unverified consumer reports, and the reported caller ID may have been spoofed. Their presence does not prove fraud; absence only describes the available dataset/window. Number-plan checks do not establish assignment, reachability, ownership or identity. The official-contact directory is limited and dated; even an exact published-number match does not authenticate an incoming caller, and a mismatch does not prove fraud. No carrier, subscriber-name, commercial phone reputation or private-person enrichment service is active in this version. Public-source corrections may require contacting the original publisher; you may also contact us about information we hold.
External evidence is checked after your deliberate submission and may be cached as public source data to reduce repeated requests. We do not save the surrounding message, selected phone input, private check history or result in an evidence-content database. Hosting/provider logs, quota metadata, the public datasets, optional AI and reports remain separate. Online evidence has its own shared capacity limit and can be unavailable even when local checking works.
Technical information and capacity
OpenAI Sites hosts this service using Cloudflare Workers and D1. Hosting/network systems process information needed to deliver requests and protect the service. Hosting operational logs can include network address, time and route. We have not established a specific Sites log-retention period, processing location or contractual zero-retention arrangement; we do not promise one.
To enforce limits, D1 stores a keyed hash derived from a network address, quota category, time-window start, request count and expiry timestamp, plus shared AI-attempt and online-evidence counters. These records do not contain submitted messages, screenshots, phone numbers, context, results or the original network address. A keyed hash is pseudonymous technical data, not a promise of anonymity. People sharing a network may share an allowance. AI and evidence capacity are shared across the beta, not guaranteed per-person allowances; failed attempts may count.
By default, quota rows become eligible for removal one day after their usage window ends. Later checks remove limited batches of expired older records from the same category while protecting active windows. Inactivity or a backlog can delay removal. An expiry timestamp is not a guaranteed deletion time. Hosting and AI-provider records follow their separate rules.
Your web page, browser and choices
Your current input, selected number, context, image preview and result stay in the open page. Changing content, context, region or review mode invalidates the earlier result and requires fresh sharing choices. Clear removes the page’s current input/result and stops waiting. It cannot recall a submitted request, undo server/provider processing, erase quota records or delete provider logs. Browser history/session restoration, clipboard contents, downloads and copies you share are under separate control.
Shared web integrations can stage text in a URL fragment, which the checker removes from navigation before parsing. Nothing is checked automatically. Browser/device software may keep its own records. We include no advertising trackers or third-party marketing analytics. Any hosting access controls and necessary cookies are separate from a product check history.
Your choices in native apps
A fresh normal launch starts with an empty review. A deliberate Share import or returning to a retained open draft can show text. It does not automatically read the clipboard, monitor messages or contacts, scan your photo library, visit links, or submit a check. You deliberately type, tap Paste, select one image, or share an item into an editable review. The English text-pattern check runs on your device. It is not AI, cannot authenticate a sender or establish safety, and can miss scams or warn about legitimate content.
The app does not implement a saved review history. The open review, local result, sharing choices and unsent report draft are temporary. Editing or clearing invalidates the old result and choices and stops waiting for pending work. Closing the review or process can discard the draft. Clear does not delete your original image, erase the operating system's clipboard or copies in other apps, recall sent data, or guarantee secure erasure of every memory buffer. Android app backup and device-transfer exclusions do not control records kept independently by the operating system, keyboard or selected document provider.
Selected screenshot text reading in native apps
Native screenshot text reading is on device: Apple's Vision on iOS and bundled Google ML Kit Latin text recognition on Android. Text extraction can omit or change words. Read and redact the extracted text before choosing a check. The native AI client sends reviewed text only, not screenshot pixels. This differs from the web screenshot feature, which uses OpenAI after a separate choice.
The iOS system photo picker and Share panel grant the selected item without broad photo-library access. Import makes a private temporary copy, capped at 12,000,000 bytes, rejects images over 40 million pixels, and downsamples for text reading. Normal completion or cancellation deletes that private copy. Abrupt termination can leave a copy until system cleanup or later app/Share startup or import performs limited stale-file cleanup. That cleanup depends on later activity; no fixed deletion time is promised. The original image is not deleted.
Android uses the system document picker or an item deliberately shared to the app, without broad storage, photo-library or camera permission. The import path uses a bounded in-memory read, rejects more than 12,000,000 compressed bytes or 40 million pixels, and downsamples and corrects orientation. It does not create an app-managed image copy on disk or keep a persistent permission to reopen the document. The app clears its temporary read buffer and returned compressed-byte array and releases bitmaps. Other app, operating-system and SDK copies are not guaranteed erased. Your document provider can have its own storage or network behavior.
Google's ML Kit may send diagnostic and usage information separately from on-device image/text processing. Google's disclosure includes a per-installation identifier and information about the device, app and configuration, performance, events and errors, and input/output sizes.
The Android app creates its text-recognition client in a deliberate image-import flow, after basic attachment checks and before reading and decoding the image. The installed app still includes ML Kit startup providers and transport components, so delaying that client does not establish that SDK technical processing or network activity waits for image selection. AI, lookup and report-sharing choices do not switch SDK diagnostics on or off, and this app offers no diagnostics opt-out. We have not independently measured SDK network traffic or the exact timing of transmissions. Android has internet permission for optional online features and SDK networking. The whole app is not promised to be telemetry-free or to make no network connections. Google's disclosure is at https://developers.google.com/ml-kit/android-data-disclosure . We have not established a specific SDK diagnostic-retention period. The app includes no advertising SDK or app-managed analytics of review content. System and SDK records remain separate from the app's lack of saved review history.
Optional online actions
Choosing to check online availability sends no review text, selected number or image, but the hosting infrastructure sees connection and ordinary technical information. Availability retrieves the current policy version, source-origin list and independent configured feature flags for combined review, evidence, AI and reporting. It does not run an AI review, prove provider health, reserve capacity or confirm consent. Each feature can be unavailable separately; reporting and non-AI evidence do not require AI to be enabled. No AI, evidence lookup or report is sent automatically by importing content.
Optional AI requires current eligibility, terms, OddFlag-processing and OpenAI-sharing choices for the reviewed content and context. Online domain/phone evidence has a separate lookup choice; it does not require AI-sharing consent. These choices are used for that request and reset when relevant content changes. Native apps contain no provider secret and require no OddFlag account or browser sign-in. The provider, hosting, quota and retention disclosures elsewhere apply to optional native online requests. Native local OCR and pattern/structural checks consume no server check, AI or report allowance. AI and evidence capacity are separately shared across web and apps; reports have separate limits and consume no AI attempt. On-device checking does not establish zero retention for an optional upload.
iOS Share and support
The iOS Share panel performs local text reading and pattern review only. Online AI, online reporting and external policy links are in the main app. To continue there, separately choose Copy reviewed text, tap Done, open OddFlag AI and tap Paste. Copy uses the local clipboard, disables cross-device copying and requests a ten-minute system expiration. It cannot recall a copy another app has already received. iOS technical support opens a mail draft with static instructions and the local review version; the review and result are not automatically attached. You control any additions and whether to send. Support correspondence follows the email-provider and retention disclosures below. The home-screen widget opens the main app; it does not automatically read or check content.
Reports you choose to send
When a result offers reporting, you can choose an issue category and optionally type a redacted note of up to 500 characters (500 UTF-16 units). A separate report choice is required before sending. The reporting feature sends it to Fifteen-Twenty-Seven LLC for review; it is not automatically sent for AI analysis or emailed to anyone. The original message, screenshot, result text and your contact details are not attached automatically. Do not paste them into the optional note; describe only the issue and leave out secrets and personal information.
Reports are stored separately in the application’s D1 database. They contain a random reference, category, platform, app/result versions, review method, receipt/expiry times, accepted terms version and any note you chose to type. A note may contain personal information if you include it. We use reports to investigate problems and improve the service; no report is automatically submitted to the AI provider. You may contact support separately with your reference if needed. Reporting does not promise an individual reply, ongoing monitoring, a payment or automatic contact.
Native reporting requires the current eligibility and terms choices and a separate report-sharing choice. It does not require the AI-sharing checkbox. The service stores the accepted terms version with the report, not a verified age/location or a full history of checkbox events. A validated receipt confirms receipt, not resolution or an individual reply. Changing the reason or note requires a fresh report-sharing choice. Do not put passwords, codes, full account or card details, identity documents or unnecessary personal information in a review or report.
Reports become eligible for deletion 30 days after receipt. Later accepted reports trigger removal of limited batches of expired reports; inactivity or a backlog can delay removal, so no exact deletion time is promised. Expiry or an operator deletion removes the report from the active application database; it does not promise immediate erasure from provider backups or recovery history. Sites-specific backup retention/deletion controls have not been verified. Usage-limit metadata and hosting records are separate. Clear removes an unsent report draft and stops waiting, but cannot recall a submitted report or delete a report already received. Keep the reference if you may want a follow-up privacy request. To request help locating or deleting a received report, contact us with its reference. The report form collects no identity/contact field to make an otherwise unidentified old report searchable by your name.
Support and privacy requests
Email support@oddflag.ai for support, privacy or security concerns. Namecheap forwards this address to the company’s Gmail. Those providers handle contact addresses, correspondence and attachments; replies may come from the company account rather than the alias. Contacting support is separate from a check or an in-app report, and no original submission is attached automatically. Send the issue category, product version and approximate time first. Redact any material you choose to share; do not send secrets.
We use correspondence to respond, investigate reported problems and handle necessary records. Email can persist in forwarding/Gmail services and is not erased by Clear. We do not promise a fixed email-deletion deadline or immediate response. Records may be needed to resolve a request, address a dispute, protect the service or comply with law. Any internal case queue uses limited category/status/role/version/time/resolution metadata rather than original messages, screenshots, contacts or arbitrary notes.
You may ask to access, correct or delete personal information we hold by emailing us. We may request the minimum needed to verify and locate a record, never passwords or security codes. Depending on applicable law, additional rights or an appeal may be available; we will explain our response and any applicable appeal route. We do not penalize protected privacy requests. Because no check history is saved, we may be unable to retrieve an old check. That does not establish that a provider has no records, and we cannot erase copies independently saved elsewhere.
Other disclosures and legal rights
We use providers to host/secure the checker, interpret content when you choose AI and deliver support you request. They may process information outside the US. We do not sell submitted content or disclose it for targeted advertising. Information may be disclosed as required by law or valid process, or when reasonably needed to address fraud, security incidents or threats to rights/safety, subject to applicable law. We do not file scam reports or contact a bank or sender for you.
The beta does not sell subscriptions or collect payment-card details. Before a paid service or materially new data use, we will update disclosures and obtain any required agreement or new choice. The age/location choice is your declaration, not independent age/location verification; do not submit children’s personal information. Contact us if you believe a child’s information was submitted.
Security and changes
We minimize deliberate storage and use protective controls, but no transmission, provider or device is guaranteed secure. Avoid unnecessary sensitive material, even in a security report. We will update this notice and its effective date when practices change and provide notice or a new choice as required. Read the terms and method and limits before checking.