Notion leaks email addresses of all editors of any public page
TL;DR Highlight
Notion exposed editor names, photos, and emails via page metadata for five years.
Who Should Read
Developers or IT administrators at companies using Notion for team collaboration and operating public pages, or where employees contribute to editing. Teams prioritizing security should immediately verify for email address exposure due to potential spam, phishing, and OSINT attacks.
Core Mechanics
- When a Notion page is published to the web, the names, profile pictures, and email addresses of all editors are automatically included in the page's HTML metadata.
- This behavior was documented in Notion's official documentation (help/public-pages-and-web-publishing) as a small footnote, making it difficult for users to discover.
- A Notion employee named Max commented on Hacker News, admitting that while the issue is documented and a warning is displayed upon publishing, it is insufficient. They are considering removing PII from the public API endpoint or implementing an email proxy system like GitHub.
- The issue has reportedly existed for at least five years, with one Hacker News user sharing their experience of being deanonymized through their Notion public page five years ago.
- Notion clarified that a fix is not as simple as a 'one-minute change,' suggesting underlying data structure or API compatibility issues.
- Some argue this isn't a Notion-specific problem, but a fundamental limitation of architectures that centrally store user data. One user suggested storing data per user as a solution, but acknowledged challenges with group collaboration, offline access, and scraping prevention.
Evidence
- "Notion employee Max officially acknowledged the issue in a comment on Hacker News, stating that documentation and a warning are present but insufficient, and that they are exploring removing PII or implementing an email proxy. A Hacker News user shared a five-year-old experience of being deanonymized through a Notion public page, demonstrating the issue's longevity. Community discussion highlighted the discrepancy between the existing documentation and its discoverability, with most agreeing that a hidden notice is insufficient. A Notion employee refuted claims of a simple fix, hinting at underlying API or data structure complexities. Some users expressed disappointment with Notion's shift towards an 'AI all-in-one app,' criticizing the prioritization of features over security and privacy."
How to Apply
- If you currently operate Notion public pages, immediately inspect the HTML source or metadata to verify whether editors' email addresses are exposed. If found, temporarily make the page private or separate editors into dedicated workspaces.
- If Notion pages edited by company employees are publicly accessible, their company email addresses could be collected for spear phishing or spam attacks. Audit your public page list and revert any unnecessary public settings.
- If you are developing services using the Notion API or public pages, re-examine the user information fields in the API response and ensure your code does not expose this information. Monitor Notion's change logs for potential updates to the email proxy system.
Terminology
Related Papers
Did Claude increase bugs in rsync?
rsync 프로젝트에 Claude AI가 도입된 이후 버그가 늘었다는 소셜 미디어 주장을 실제 데이터와 통계 분석으로 검증한 글로, 결론적으로 Claude 도입 후 릴리즈가 역사적 분포에서 유독 버그가 많다는 통계적 근거는 없었다.
I built a vulnerable app and spent $1,500 seeing if LLMs could hack it
Firebase 취약점을 가진 앱을 직접 제작하고 GPT-5.5, Claude, Deepseek 등 주요 LLM이 자율적으로 해킹할 수 있는지 실험한 결과, GPT-5.5가 70% 성공률로 압도적이었고 Claude는 보안 거부 정책 때문에 능력과 무관하게 낮은 점수를 기록했다.
Clustered Self-Assessment: A Simple yet Effective Method for Uncertainty Quantification in Large Language Models
LLM이 여러 답변을 의미 단위로 묶어 객관식으로 만들고 스스로 채점해서 '이 답 얼마나 확신해?'를 수치로 뽑아내는 기법.
SkillHarm: Lifecycle-Aware Skill-Based Attacks via Automated Construction
AI 에이전트가 사용하는 'Skill 패키지'에 악성 페이로드를 심으면 최신 모델도 86%까지 뚫린다는 보안 벤치마크.
MemTrace: Tracing and Attributing Errors in Large Language Model Memory Systems
RAG, Mem0 같은 LLM 메모리 시스템이 왜 틀린 답을 내는지 자동으로 찾아주는 디버깅 프레임워크
DeepSWE: A contamination-free benchmark for long-horizon coding agents
기존 SWE-bench의 데이터 오염 및 검증 오류 문제를 해결하기 위해 처음부터 새로 만든 코딩 에이전트 벤치마크로, GPT-5.5가 70%로 1위를 차지하고 모델 간 성능 격차가 훨씬 뚜렷하게 드러난다.