| --- |
| license: other |
| license_name: research-use-only |
| license_link: LICENSE |
| language: |
| - en |
| pretty_name: APT Reports Collection |
| size_categories: |
| - 1K<n<10K |
| tags: |
| - cyber-threat-intelligence |
| - apt |
| - security |
| - threat-reports |
| --- |
| |
| # APT reports |
|
|
| > Collected with [CTI-Crawler](https://github.com/peng-gao-lab/cti-crawler) in APT mode. The full collection is the ZIP file `apt_github_distribution_20260922T211548298793Z.zip` (about 10.7 GB); the record files next to it are copies of the ZIP's root records so they can be read without downloading the archive. |
| |
| This collection contains 3051 articles, with 2364 HTML files and 3051 PDFs (687 downloaded as PDF and 2364 converted from HTML). An article available in both formats is counted once. |
| |
| ## Layout |
| |
| ```text |
| 01_MITRE_references/ |
| html/ |
| pdf/ |
| 02_APT_publishers/ |
| <source>/ |
| html/ |
| pdf/ |
| 03_activity_reports/ |
| <source>/ |
| html/ |
| pdf/ |
| ``` |
| Files use short article numbers, such as `000081.html` and `000081.pdf`. Both versions of an article share the same number. |
| |
| The article number is assigned when the article is collected and never changes. `filename_map.json` links each number to its source, URL and crawler output path. Entries with `included: false` would identify PDFs that could not be generated. In this release every HTML report has a PDF, although some PDFs have issues documented in `pdf_issues.json`. Number order is not collection order. |
| |
| ## Records |
| |
| | File | Contents | |
| | --- | --- | |
| | `classifications.json` | Article URLs, sources, collection categories, and paths using original filenames. | |
| | `filename_map.json` | Article numbers with their source, URL, crawler output path and path in this archive. | |
| | `source_risks.json` | Second-category sources with more than 20% non-APT material in the reviewed sample. Microsoft is also flagged because of its broader content mix, although its full-text sample did not exceed 20%. Reasons and examples are included. | |
| | `pdf_issues.json` | PDFs with missing images or text-layer defects, with details per file. | |
| | `collection.log` | Packaging time and article counts for each source and source category. | |
|
|
| ## Review notes |
|
|
| The crawler uses three categories of sources. **The second category, `02_APT_publishers`, comes from publishers' own APT sections and series and has been checked at the source and sample level, not article by article. The first and third categories were selected for their relevance to APT research, but they have higher rates of non-APT content.** |
|
|
| `01_MITRE_references` contains documents collected from the references on MITRE ATT&CK's Groups pages. We chose these references because many of the listed groups are APT groups that have carried out APT activity. We download all eligible references found in the crawl. They are not all APT activity reports. |
|
|
| `03_activity_reports` contains material collected directly from publisher websites and third-party repositories. These sources contain substantial APT-related material, but they have not undergone the same sample-based review as the second category. |
|
|
| - ESET Research and Seqrite Technical reports were collected from the publishers' websites. These sections include APT activity, other attacks, malware analysis, and broader security research. |
| - `cybermonitor_history` contains PDFs downloaded from [CyberMonitor's report collection](https://github.com/CyberMonitor/APT_CyberCriminal_Campagin_Collections). |
| - `blackorbird_reports` contains PDFs downloaded directly from [blackorbird/APT_REPORT](https://github.com/blackorbird/APT_REPORT). We selected its actor and campaign directories plus one reviewed BfV/BSI Signal advisory, and excluded 14 reports confirmed to be already held in our collection. |
|
|
| `02_APT_publishers` contains sources with more relevant APT reports. But our sample checks still found some non-APT content. |
|
|
| All HTML reports have corresponding PDFs. 392 of them are missing one or more images that could not be retrieved (placeholders mark the spots). 2 PDFs display their text correctly but lose a few combining marks of a complex script (for example Myanmar medials in a decoy file name) when the text is copied or searched. `pdf_issues.json` lists the affected characters and text, and the HTML originals keep them. See `pdf_issues.json` for the full list and details. |
|
|
| ## License |
|
|
| The reports in this collection remain the copyright of their original publishers. We only redistribute copies for non-commercial academic and security research. The index and record files we generated (README, classifications.json, filename_map.json, pdf_issues.json, source_risks.json, collection.log) are released under CC BY 4.0. Rights holders who want a report removed can contact changzeli9066@gmail.com. |
| |
