文档与截图:分享前移除敏感数据

黑色矩形可能只是遮住信息,并未删除内容。制作分享副本,然后验证接收者实际拿到的文件。

虚构的无人文档审查工作台,放有两个黑色文件夹、发光检查面板和合上的笔记本电脑。 由人工智能生成的编辑插图

六项可重复的检查

  1. 限制内容

    确定接收者需要理解什么。在副本上操作,将原件保存在获准的位置。

  2. 查找身份信息

    检查姓名、地址、编号、浏览器标签、通知和文件名,不要只查看截图中央。

  3. 真正移除内容

    对 PDF 应用实际删除内容的涂黑处理。覆盖对象或隐藏电子表格行并不等于删除数据。

  4. 检查隐藏信息

    使用适合文件格式的工具检查作者、批注、修订和附件。导出后再次检查。

  5. 验证最终副本

    重新打开导出文件,尝试搜索及复制应被删除的文字,检查所有页面和可读性。

  6. 检查分享权限

    核对接收者、权限和访问期限。在未登录状态下测试链接,记录批准版本,但不要重复已删除的数据。

需要保存的验收记录

  • 检查的文件及版本
  • 已移除的数据类型
  • 对导出文件的测试
  • 接收者、权限及复核人

这是工作方法,不是符合性证明。请调整范围并记录例外。

不同情况与适用检查

小屏幕上可横向滚动表格。使用键盘时,将焦点移到表格并使用方向键。

情况说明的问题有效检查
黑色矩形遮住姓名,但搜索仍能找到文字。视觉遮盖可能保留底层数据。执行真正删除数据的密文处理并检查导出文件。
预览正常,但仍有批注或附件。可见页面以外仍存在信息。检查原文件以及所有用于共享的导出文件。
链接无需登录即可访问。访问范围可能超过预定收件人。限制访问并在退出登录后测试链接。
审批后又修改了文档。原审批针对的是其他版本。重新检查最终版本并明确记录获批文件。
隐藏行参与工作簿公式。删除它们可能改变保留的计算结果。在副本中显示隐藏行,检查依赖关系并核对清理后的结果。

虚构示例

一个虚构团队准备分享项目报告和截图。其中包含收件人不需要的审核人姓名和内部参考编号。

在副本上操作,删除不必要的信息与隐藏数据,然后导出。另一名审核人下载该版本,搜索已删除词语并测试文字复制。记录只描述删除的信息类别,不重复机密值。

验收标准

  • 已明确收件人、共享目的及必要信息。
  • 下载版本通过符合其格式的文字、隐藏数据和访问检查。
  • 已记录获批版本、审核人和访问条件。
  • 清除无用信息后,必要的数值和计算仍保持一致。

实际问题

模糊处理足够吗?

不要把它视为保证。应从分发副本中移除信息,并验证最终结果。

导出为 PDF 会清除所有信息吗?

不会。检查导出文件的内容与隐藏信息;不同应用的能力不同。

应删除原件吗?

遵循获准的保留政策。将受保护的原件与分享副本分开。

搜索不到结果就证明数据已删除吗?

不能。扫描件可能显示信息,却没有可搜索的文本。还应检查最终副本中的图片、所有页面及相关隐藏信息。确认搜索能找到保留的词语:搜索功能本身失效时,零结果可能造成错误的安全感。

修正版会替换旧附件吗?

不会自动替换。检查邮件、附件、共享链接及接收者可查看的历史记录。明确批准的版本,移除你能控制的旧副本。发布新版本无法收回已经收到的文件。

只看预览能批准数据删改吗?

不能。下载实际共享的文件,检查内容、搜索、文字复制及相关隐藏数据。预览可能与导出文件不同。

如何保留证据而不复制敏感数据?

记录删除的信息类别、检查项目、版本和审核人。不要在截图或广泛可见的工单中重复已删除的值。

官方参考资料

参考资料:数据、用途与局限

文档核查日期 .

此语言版本的详细参考说明以英语提供。

参考资料:数据、用途与局限
来源与目的已核实的数据或指标使用方法需要保留的局限
C2PA · Content Credentials

Describe and verify certain content provenance information.

C2PA specification 2.4 defines manifests, assertions and cryptographic binding mechanisms associated with media. Examine what the manifest asserts, who signed it and whether validation succeeds. Test preservation after transformations and distribution.

深入了解方法 →

Validated provenance does not establish the truth of a scene. Absence of a manifest does not demonstrate a fake.
NIST · Privacy Framework

Examine risks to people arising from data processing.

The Privacy Framework is a voluntary privacy risk-management tool. It is distinct from a universal legal compliance checklist. Describe actual processing: data, purposes, recipients, access and retention. Compare documents, settings and deletion or retrieval tests.

深入了解方法 →

Using the framework alone does not determine compliance with applicable law. Processing context and jurisdiction still need examination.

STIX 指标与 TLP 共享

此语言版本的详细参考说明以英语提供。

Check TLP scope before publishing an excerpt

A traceable note does not by itself grant permission to redistribute it. TLP markings, additional restrictions and intended recipients must accompany a redaction or publication decision.

  • AMBER+STRICT limits sharing to the organisation; AMBER’s scope for protecting clients does not apply by default.
  • GREEN excludes public channels; CLEAR does not remove copyright rules or separate restrictions.
  • Check the marking of each quoted field and obtain explicit source permission if the intended sharing exceeds the received scope.

Read the STIX, observations and TLP sharing reference →

Sources checked on · FIRST · TLP 2.0

PDF:内容删改与分发

此语言版本的详细参考说明以英语提供。

Check the delivered copy beyond its blacked-out areas

Redaction aims to remove excluded information from the copy given to the recipient. Review must cover the final file and its access paths: visual masking, a search returning no results or metadata cleanup alone does not cover the whole decision.

This is an editorial acceptance-review method, not a PDF analysis tool. The five versions and forty observations are fictional; this page uploads, sanitizes and certifies no files. Cited Adobe features concern Acrobat Pro. ICO guidance was published in a UK context and is marked under review; it is not presented as French legal validation.

Four operations and their boundaries

Four operations and their boundaries
OperationIntended outcomeRemaining checks
Mark an areaIdentify content for removal in the appropriate tool.A pending mark or overlaid rectangle is not an applied, saved redaction.
Apply redactionRemove targeted text or images from the copy using the intended function, then save.Hidden and embedded content needs separate inspection; the initial selection may be incomplete.
Sanitize hidden informationInspect and remove relevant metadata, comments, layers or embedded content.Check the saved copy and software limits. Do not assume PDF export removes every item from the source document.
Approve releaseBind a precisely identified version to recipients and documented checks.Links, attachments and accessible prior versions may expose other files. Review remains limited to its scope.

Eight surfaces to examine before handover

Eight surfaces to examine before handover
SurfaceReview questionUseful evidence and limit
Visible textWere intended passages actually removed from the saved copy?Review all pages and test copying the relevant text. A black appearance does not establish removal.
Images and OCR layerDoes information removed from text remain in an image, or vice versa?Inspect the image and text layer where present. A scan without searchable text may still display the information.
Comments and annotationsDo remarks, names or references remain accessible in review panels?Inventory relevant annotations and verify the cleaned version intended for sharing.
Metadata and propertiesDo author, title, keywords or custom properties reveal excluded information?Inspect relevant categories using a suitable tool. Removing a property does not establish absence of the value elsewhere.
Attachments and embedded objectsDoes an attached file or object contain another version of removed material?Review the inventory and every retained item. The visible page does not necessarily describe all embedded content.
Layers and non-visible objectsDoes the file contain elements omitted by an ordinary preview?Use available software controls and record uncovered areas. An absent warning is not a complete review.
Office source documentDo changes, embedded data or properties reappear in the export?Work on a copy; Document Inspector depends on format and cannot detect everything. Inspect the final PDF again.
Release containerDoes the message, shared folder or version history expose an old file?Check the file obtained by the intended recipient and accessible versions. Correcting one copy does not erase copies already received.

Forty observations without a security score

Fictional exercise: FICTIF-42 must be removed, while PUBLIC-ALPHA must remain present and searchable in these five test versions. No real confidential data is used. States describe acceptance observations already stipulated in the scenario; they are not calculated from a PDF.

SatisfactoryObserved failureNot checked
Forty observations without a security score
Scenario checkR01R02R03R04R05
Targeted areas renderedSatisfactorySatisfactorySatisfactorySatisfactorySatisfactory
Retained control PUBLIC-ALPHA foundSatisfactoryObserved failureSatisfactorySatisfactorySatisfactory
Removed term FICTIF-42 not foundObserved failureSatisfactorySatisfactorySatisfactorySatisfactory
Images and OCR reviewedNot checkedSatisfactorySatisfactorySatisfactorySatisfactory
Hidden content reviewedNot checkedSatisfactorySatisfactorySatisfactorySatisfactory
Delivered file = reviewed versionSatisfactorySatisfactorySatisfactoryObserved failureSatisfactory
Recipient access checkedNot checkedSatisfactorySatisfactorySatisfactorySatisfactory
Peer review closedNot checkedNot checkedSatisfactoryNot checkedNot checked

Interpret a negative search only with its limits and a suitable positive control. In R02, PUBLIC-ALPHA is not found: the expected search test fails even though FICTIF-42 is not found either. “Not checked” never counts as success; an average cannot compensate for a failure.

Five decisions bounded by the fictional dossier

Five decisions bounded by the fictional dossier
VersionCandidate copy / delivered fileScenario decision
R01training-report-R01.pdf / training-report-R01.pdfHold and correct
R02training-report-R02.pdf / training-report-R02.pdfHold and correct
R03training-report-R03.pdf / training-report-R03.pdfSubmit for release decision within the tested scope
R04training-report-R04.pdf / training-report-R02.pdfHold and correct
R05training-report-R05.pdf / training-report-R05.pdfReview still open
  • R01: the removed term is still found and several checks remain open.
  • R02: the positive control fails. No search result alone does not demonstrate redaction.
  • R03: the eight defined checks are satisfactory in this exercise. This guarantees neither exhaustive search nor absence of risks outside scope.
  • R04: R02 is delivered although R04 was reviewed. Acceptance does not transfer to the wrong file.
  • R05: defined technical checks are satisfactory; peer review is not closed. The decision remains open.

Eight review situations to explain

01 · Visible rectangle, text still copyable

Fictional situation: layout shows a black area but selection or extraction recovers the passage.

Decision: repeat redaction with a suitable function, apply it, save a copy and inspect the result. The signal shows removal has not been established; changing rectangle colour does not fix it.

02 · Negative search without a positive control

Fictional situation: neither the removed term nor the expected retained term can be found.

Decision: suspend the search conclusion. Check whether the tool and text layer suit this document, then add image inspection. The negative test alone does not establish removal.

03 · Search-clean scan, information visible in the image

Fictional situation: search cannot find a name on a scan with no usable text layer, but the image still shows it.

Decision: handle the information in its actual representation and review the final copy. Missing OCR is not proof of absent visible information.

04 · Comments removed, attachment overlooked

Fictional situation: the comments panel is empty; an embedded attachment still contains an old version.

Decision: examine the attachment and its purpose for the recipient, then correct the release copy. Passing comment review does not cover embedded objects.

05 · Metadata cleaned, name still visible on the page

Fictional situation: the author property is blank but a name remains in a footer.

Decision: revisit the visible-content inventory. Property cleanup and passage removal are separate checks; neither validates the other.

06 · Correct version approved, wrong attachment sent

Fictional situation: review concerns R04; the prepared message contains training-report-R02.pdf.

Decision: do not release the message as approved. Bind the final version to the attachment or file actually obtained by the recipient, then record a fresh check.

07 · Image conversion loses accessibility

Fictional situation: conversion removes useful text and makes the file difficult to read with assistive technology.

Decision: assess the method and produce a usable copy without reintroducing excluded information. Conversion is no universal guarantee; test its effect on readability, search and accessibility.

08 · Corrected copy, old version still accessible

Fictional situation: the corrected PDF is online but a previous version remains downloadable from the shared folder.

Decision: also address access and versions under your control, then assess prior disclosure. A new version does not automatically recall files already copied or received.

Six acceptance records without repeating removed values

  • Sharing purpose, recipients and information categories to remove or retain.
  • Reviewed-copy identity, date, tool, version and relevant processing settings.
  • Surfaces covered, positive and negative checks, results and coverage limits.
  • Corrected discrepancies and review of relevant hidden or embedded content.
  • Link between the reviewed copy, final attachment and versions accessible to the recipient.
  • Reviewer, decision, authorised scope and prior-copy follow-up under the applicable policy.

Forty cells, identifiers and decisions are a local teaching construction. Sources explain the checks; they attest to no file analysis performed here.

Product sources and methodological guidance

Sources reviewed on