AIT OSINT 2025 Data Set

The cyber security news items in this dataset were gathered from publicly accessible sources using the OSINT platform Taranis AI  (https://taranis.ai/). The dataset spans the period from January 01, 2025, to December 31, 2025, and consists of daily JSON files containing a total of 55.746 news items in either German or English.

Metadata
Taranis AI automatically collects news items in eight-hour intervals and processes them for further use. During this process, the content, author, publication date, and title are directly sourced from the RSS feed without modification. The platform then adds the following attributes to the news item: An ID, the link to the collected news item, the ID and URL of the open source, and a hash generated from the author, title and link.

Furthermore, each news item in the dataset was assigned content-related tags using three large language models (LLMs): gpt-oss-120b, Qwen3.6-35B-A3B, and gemma-4-31B-it. Indicators of compromise (IOCs) were additionally extracted from the news items using the Python library ioc-finder. Finally, the language of each news item was determined using Python’s langdetect library and encoded according to the ISO 639-1 standard.

Each news item consists of

• id-- ID of news item (Format: UUID)
• published-- Timestamp of news item's publication (Format: Timestamp)
• author -- Author of collected news item (Format: String)
• title -- News item’s title  (Format: String)
• language -- News item's language (Format: String)
• link -- Link to news item (Format: URL)
• content -- News item's content (Format: String)
• hash -- SHA-256 hash of following attributes: Author, title, and link (Format: SHA-256)
• osint_source_id -- ID of open source (Format: UUID)
• osint_source_url -- URL of open sourc (Format: URL)
• tags -- Tags representing news item’s content (Format: Nested JSON)

The associated pdf in this data set provides further information on the number of news items collected per month, as well as distribution across languages, most common sources, and tags. It also contains of a full list of public sources used for the collection.

If you use the AIT OSINT 2025 data set, please cite the following publications:

[1] Skopik, F., Akhras, B., Woisetschläger, E., Andresel, M., Wurzenberger, M., Landauer, M. (2024). On the Application of Natural Language Processing for Advanced OSINT Analysis in Cyber Defence. In Proceedings of the 19th International Conference on Availability, Reliability and Security (pp. 1-10).

[2] Woisetschläger E., Andresel M., Skopik F., Akhras B., Leitmann P., Landauer M., Wurzenberger M., Schindler A. (2025). SC4OSINT: A Story Clustering Approach to Optimize OSINT Analysis. In Proceedings of the 20th International Conference on Availability, Reliability and Security (ARES) (pp.5-24)