Use this as a triage index. Start with the pattern name and symptom, then open the audit details only when the issue looks relevant.
Use the counts to decide which risk area to review first.
Common symptoms that usually deserve immediate review.
AI agents can execute real-world side effects before users understand or authorize them, especially when plan previews, progress traces, chat messages, or notifications blur proposed actions with completed actions.
Open diagnosisUsers receive a confident-looking AI answer that includes factual claims or recommendations but no usable source, source scope, retrieval status, unsupported-claim label, or verification path.
Open diagnosisUsers over-trust AI output when the interface presents raw, uncalibrated, stale, or out-of-scope model scores as exact confidence percentages, decimals, probabilities, or certainty levels.
Open diagnosisUsers can commit destructive or high-consequence actions because the final labels do not say the real outcome, target, scope, or safe alternative.
Open diagnosisOpen a row when you need detection cues, remediation, and safer alternatives.
AI agents can execute real-world side effects before users understand or authorize them, especially when plan previews, progress traces, chat messages, or notifications blur proposed actions with completed actions.
The agent can call tools, send external messages, spend money, issue refunds, change access, update customer records, deploy code, delete data, submit forms, publish content, or trigger downstream workflows.
Users receive a confident-looking AI answer that includes factual claims or recommendations but no usable source, source scope, retrieval status, unsupported-claim label, or verification path.
The answer may be generated from web search, file search, enterprise knowledge sources, selected documents, uploaded files, chat history, model prior knowledge, tool outputs, or no retrieval at all.
Users over-trust AI output when the interface presents raw, uncalibrated, stale, or out-of-scope model scores as exact confidence percentages, decimals, probabilities, or certainty levels.
The UI displays model output, retrieval ranking, classifier score, extractor score, recommendation rank, generated-answer score, or risk estimate as if it were a calibrated probability.
Auto-advancing carousels without a reachable pause or stop path interrupt reading, hide controls, move focused or hovered content, create noisy announcements, and can make users miss critical information.
A carousel, hero, promotion rail, onboarding slider, or recommendation strip changes visible content on a timer.
A drawer opens from the current page but lacks a reliable way to dismiss it and return to the exact context that launched it.
A side drawer, inspector, side sheet, mobile full-screen drawer, or navigation drawer opens over or beside the current page.
Users must guess what an icon-only control does, which slows action selection, hides risk, and can leave keyboard and assistive technology users without an action name.
A toolbar, table row, card, or compact navigation surface contains icon-only controls.
A page presents routine information in a modal dialog even though users do not need to stop the current task, make a blocking decision, or protect background state.
The layer contains read-only help, status, preview, metadata, release notes, success feedback, onboarding tips, marketing, or supplemental explanation.
A form, checkout, application, account change, or decision flow needs required instructions, but the only copy lives in a temporary tooltip that can disappear before users can apply it.
The hidden content is needed to complete a field, understand eligibility, choose an option, avoid a fee, meet a deadline, satisfy a document rule, or recover from a blocked control.
Users can commit destructive or high-consequence actions because the final labels do not say the real outcome, target, scope, or safe alternative.
A dialog, sheet, action menu, toast, command palette, mobile action sheet, or review page includes a destructive command.
When products ask users to confirm too many routine actions, users learn that confirmations are meaningless friction. The prompt no longer creates attention when the consequence is genuinely destructive, costly, external, or hard to undo.
A product uses modal confirmations, alert dialogs, browser confirms, action sheets, or extra OK prompts for many commands in the same workflow.
Users encounter disabled buttons, menu items, toggles, form controls, or workflow actions that block progress without a usable route to meet the requirement, request access, resolve a dependency, recover from state, or choose a safe alternate path.
The interface gates an action behind form completion, permission, quota, account status, dependency setup, offline state, session expiry, review approval, or safety policy.
Undo creates strong user trust, but products sometimes expose an undo label before they can actually reverse the action. The result is worse than no recovery because users believe a harmful change was undone when data, permissions, order, messages, or external effects remain changed.
A product shows Undo in a toast, snackbar, activity row, command bar, keyboard shortcut, history stack, or recovery panel after a completed action.
Validation is supposed to help users correct data, but clearing input after errors turns a small correction into rework, loses context, and can make users abandon or mistrust the form.
A form, editor, checkout, upload, import, authentication, or configuration screen validates user-entered values on blur, submit, save, retry, or server response.
Users reach a blank or nearly blank area and cannot tell what is missing, why it is missing, or what they can do next.
A list, table, dashboard, workspace, inbox, or panel can render with no visible items.
Users see an unavailable action but cannot tell which field, permission, prerequisite, system state, or safety rule blocks it.
A form, setup flow, permission model, quota, dependency, or safety gate controls whether an action can run.
Users are left in an unbounded busy state when a request, save, import, sync, or report generation may never complete and the UI offers no timeout, fallback, cancellation, retry, or failure explanation.
A request can hang, fail silently, be queued for a long time, or lose its completion event.
A blocking or high-consequence failure is announced only in a transient toast, so users can miss what failed and lose the path to recover.
A payment, save, permission, destructive action, security, or data-integrity operation fails.
A consequential transaction completes, but the only success feedback is a transient toast that can disappear before users can verify the outcome, copy proof, understand next steps, or recover the receipt.
The user has completed or appears to have completed an application, booking, order, payment, account change, survey response, publication, approval, or service request.
Conditional form logic often hides fields to reduce noise, but if hidden fields remain required, stale, submitted, or server-enforced without a visible recovery path, users cannot tell what blocks completion.
A form has radio, checkbox, select, eligibility, account type, permission, or saved-draft answers that control which later answers are required.
Users need to loosen or remove filters, but the reset action silently clears unrelated search or result state that should remain independent.
The result set combines keyword search, filter criteria, sort, scope, saved search, saved filter, pagination, view density, or layout state.
Automatic infinite loading can make the page bottom unreachable, preventing users from reaching footer utilities, legal links, support routes, feedback, language controls, or a reliable end state.
The page uses infinite scroll, auto-load-on-near-bottom, virtualized lists, dynamic feeds, product grids, search results, activity streams, or media streams.
A visual dropdown looks like a select control but lacks the semantics, keyboard support, selected state, focus behavior, or form value exposure users expect.
A team replaces native select styling with divs, popovers, portals, or JavaScript-only menus.
Consent decisions become untrustworthy when the interface steers users toward acceptance through unequal visual weight, more steps to refuse, hidden decline, preselected options, bundled purposes, consent walls, repeated nags, misleading legal bundling, or optional processing that starts before the user chooses.
The surface asks for optional data use, marketing, research contact, AI training, partner sharing, personalization, sensitive-data use, non-essential cookies, local storage, advertising tags, analytics tags, or similar tracking.
Account deletion becomes deceptive or unsafe when the product hides the deletion path, substitutes deactivation or export, requires vague support contact without status, omits a web route after uninstall, or commits account-level destruction without naming account identity, affected data, retained data, billing, linked apps, and recovery limits.
The product lets users create or authenticate an account that owns personal data, app data, profile data, messages, files, subscriptions, linked apps, child-account data, enterprise memberships, or authentication tokens.
Users are more likely to deny, distrust, or misunderstand permission requests when a product asks before the relevant feature is visible, uses vague benefit copy, requests the wrong resource, bundles multiple resources, or provides no fallback after denial.
The prompt requests an operating-system, browser, or app-level permission for a device resource or powerful feature such as location, camera, microphone, photos, contacts, notifications, Bluetooth, clipboard, motion sensors, or storage access.