English version
Terms of Use & Privacy Policy
This English version describes the same policy as the Korean text. If an interpretation differs, the Korean text controls to the extent permitted by mandatory law.
1. Terms of use
These terms apply to the Paperlayer website and reading features. Using the service subjects you to these terms. Selecting Continue with Google means that you agree to these terms and acknowledge the privacy policy below. New paid features or separate conditions will require advance notice and any consent required by law; existing use will not be charged retroactively without that process.
Paperlayer is currently offered as a public demo. In this document, “Paperlayer” and “operator” mean the operator of this service. The service currently offers no purchase, paid subscription, or refundable transaction. Contact admin@paperlayer.ai.
The current demo covers public arXiv papers and does not accept private or unpublished document uploads. If chat is available, use it only for questions about the public paper; do not enter personal, confidential, or unpublished information. Prepared AI output may be displayed immediately. Opening an uncached paper or requesting a translation or available chat may start live generation under the interface notice and privacy policy below.
Paperlayer provides AI-assisted summaries, highlights, margin notes, translations, and links back to evidence in selected research papers. These outputs and evidence links are generated or selected by AI or automated systems. Because AI and machine learning are probabilistic, output may be inaccurate, incomplete, misleading, outdated, offensive, or fail to reflect actual people, places, or facts. The same input may produce different output. Output does not represent an endorsement, warranty, or the views of Paperlayer or a paper's authors.
Automatically selected evidence may not support an output or may link to the wrong source location. Before using or sharing output, you must evaluate its accuracy and suitability for your purpose, including appropriate human review. Do not treat AI output as the sole source of truth or factual information; verify important content in the original paper, not only against the linked evidence. Output does not replace the original paper or legal, medical, financial, or other professional advice and must not be used for decisions with significant effects on a person's rights, safety, or legal status.
Papers remain the work of their authors and rights holders. Paperlayer does not claim ownership or author endorsement. For arXiv papers, the verified, version-pinned PDF is fetched from arXiv directly. A processing copy may be retained for live AI work; it is never served to users or made public. Rights holders may request removal at admin@paperlayer.ai.
Do not bypass authentication, rate limits, or security controls; disrupt the service; send abusive automated traffic; attempt unauthorized access; or infringe another person's rights.
To the extent permitted by applicable law, the service is provided "as is" and "as available." Paperlayer makes no express or implied warranty that the service will be uninterrupted, error-free, or free of harmful components; that every paper will remain available; that AI output will be accurate, reliable, complete, safe, or fit for a particular purpose; or that saved state will not be lost or changed. You must consider the limitations above and review AI output and automatically linked evidence before using or sharing it. To the extent permitted by law, considering the public-demo nature of the service and the disclosed AI limitations, Paperlayer is not liable for indirect, incidental, special, or consequential loss caused by matters the operator could not reasonably control, foresee, or prevent. This clause does not unreasonably exclude legal responsibility or shift it to the user. These limits do not apply to liability or rights that cannot lawfully be limited, including loss caused by Paperlayer's intent or gross negligence, or injury to life or body.
2. Privacy policy
The operator of Paperlayer is the controller, and Paperlayer Privacy is the privacy contact. Contact admin@paperlayer.ai to request access, correction, deletion, suspension, or withdrawal, or to ask a privacy question.
- Without login, a signed random identifier and reading state such as hidden terms provide browser-linked continuity. The essential anonymous cookie lasts up to one year. Related state remains until it is merged, deleted on request, or the service ends.
- If you choose Google login, Google account ID, email, name, profile image URL, and an account and session identifier authenticate the account. Starred-paper IDs, titles, saved times, and reading preferences are then synced across devices. Optional onboarding may also store up to three broad research-interest categories, a broad reader role, up to three usage-purpose categories, a preferred language, and whether onboarding was completed. Every answer is optional, no free-text answer is collected, and these fields are not used for marketing analysis. They remain while the requested account feature remains active. Verified deletion requests are handled as described below. Authentication cookies and tokens are also affected by logout, expiry, and the authentication provider's policy. Paperlayer requests only OpenID, email, and profile; it does not request Gmail, Drive, Calendar, offline access, or retain Google provider tokens. On first login, Paperlayer stores a record used to merge anonymous state into the account once; that record is included in an account-deletion request.
- Inquiry email, message, and handling records may remain in the operator's email until the request and necessary follow-up are complete. Unneeded copies are deleted, while provider records and backups follow that provider's policy. Records may remain longer when needed for a legal obligation or dispute.
- Live AI processing may use a public paper's title, authors, body, selected passages, requested language, and, when chat is available, the question and recent conversation. Paperlayer does not intentionally include an account ID, email, cookie, anonymous ID, or end-user IP in model requests; information typed into chat is part of the request. Shared summaries, margin notes, and translations may remain until a removal request is completed, the paper is removed, or the service ends. Paperlayer does not separately store chat messages or answers. The current browser tab keeps the conversation until refresh or close, but request content may be handled under the OpenAI terms described below. Intermediate material kept for public-paper processing is separated from user account and session data and is not made public.
- Hosting providers may process IP address, time, path, browser/device, response, and diagnostic information for delivery, security, and troubleshooting. To check service quality, Paperlayer stores only identifier-free daily aggregates of whether papers were readable, how quickly they opened, and how the page performed, and deletes them after 35 days. Raw measurements, URLs, and element details are not stored.
- To limit abusive use, Paperlayer uses an account or anonymous identifier and client IP to check request frequency. Stored values are transformed so the original cannot be recovered; the identifiers, IPs, cookies, and emails themselves are not stored. They are deleted 48 hours after last access.
- Paperlayer uses Vercel Web Analytics default page views to understand visit and page-use trends. Under Vercel's guidance, aggregated data can include the page visited, referrer, approximate location, browser, OS, device type, and visit time. The feature does not use third-party cookies. Actual retention follows the applicable Vercel plan and policy; Paperlayer does not guarantee a particular deletion time.
- If a signed-in user intentionally submits Report an issue, Paperlayer stores the paper ID, selected category, a 10–2,000-character description, whether the reporter was signed in, handling status, and handling times to investigate and track the report. The reporter's account identifier, email, IP, browser details, and current URL are not stored. Do not enter personal or confidential information. Reports and handling history currently have no automatic deletion deadline and may remain until the operator removes unneeded records or service data is retired.
- When a readable paper opens, Paperlayer aggregates, per day and per paper, the number of opens, estimated active reading time, and whether the summary, evidence, and reading settings were used. This shows which papers and reading settings receive meaningful use. Daily aggregates are retained for up to five years and then regularly deleted. Individual reading, setting-change, or click events and reading times more precise than the day are not retained, and no user/anonymous ID, email, cookie, IP address, or session ID is stored with the aggregate.
- Requested non-English translation languages and the kind of screen used (desktop or mobile) are stored as daily totals. These totals contain no paper ID, user/anonymous ID, email, cookie, IP, session ID, or individual event time, and the two are not joined to each other or to a paper. Daily totals are kept for up to five years and then regularly deleted.
Paperlayer does not use research interaction logs, advertising cookies, or per-user retention analytics. Page views and user-submitted quality reports are handled separately as described above.
3. Cookies and service providers
Essential cookies include the signed anonymous cookie (up to one year) and, after login, session cookies until logout or expiry. Blocking them can disable saved state, login, and sync but does not prevent access to public paper source links. Paperlayer does not set advertising or cross-site tracking cookies.
- Supabase, Inc. (United States and other provider operating regions): Supabase processes account and session data, anonymous reading preferences, starred-paper state, abuse-prevention request records, identifier-free daily paper-interest aggregates, and user-submitted quality reports for the periods above. Operational-log and backup retention follow the applicable plan and provider policy. Contact privacy@supabase.com.
- Vercel Inc. (United States and other provider operating regions): web hosting, delivery, security, and diagnostics. Access records and Web Analytics page views are processed as described above. Their retention and other security or legally required records follow the applicable Vercel plan and Vercel's notice.
- Google LLC (United States and Google operating regions):optional OAuth. Google sends the account fields you authorize to Paperlayer's account authentication under Google's policy. Paperlayer does not send Stars or reading preferences to Google.
- OpenAI OpCo, LLC (United States): when an uncached paper is opened or a translation or chat is requested, Paperlayer sends the public paper content and request content over an encrypted connection to generate summaries, margin notes, translations, and answers. Account, cookie, and anonymous identifiers are not included. Retention and model-improvement use follow OpenAI's policies and settings, which Paperlayer does not independently control or guarantee. OpenAI's policy
Transfers occur over encrypted connections when you visit, save state, or sign in. You can decline Google login and still read public papers. Declining essential hosting processing means Paperlayer cannot provide the web service.
To decline OpenAI processing, do not start a live AI operation on an uncached paper and use the original arXiv PDF link instead. You will not receive a new summary, translation, or chat answer, but can still use already-published reading aids and the original-paper link. If the provider or processing method changes, Paperlayer will update this section and follow any notice or consent process required by law.
4. Deletion, security, and changes
Send an account-deletion request from the linked Google email. After verification, the operator deletes the account, Stars, reading preferences, and the anonymous-state merge record from the systems it can access and reports the outcome. Authentication records, issued tokens, and provider backups may follow provider policies and technical lifecycles. Paperlayer does not promise their immediate deletion. Safeguards include transport encryption (HTTPS), secure cookie settings, signed anonymous identifiers with secret rotation, no database access exposed to the browser, request-rate limits, and access controls.
Children under 14 must not use Google login. If Paperlayer learns that such account data was collected, it will be deleted after verification. This policy is effective August 22, 2026. Policy changes update this page and the effective date, and follow any advance notice or consent process required by law.