Translation problem: Mass of intercepted material overwhelms the BND

BND interception stations record many conversations every day, intercepting emails and the like. Use of some AI translators is banned, which affects evaluation.

listen Print view
Logo of the Federal Intelligence Service

(Image: nitpicker/Shutterstock.com)

3 min. read

The German Federal Intelligence Service (BND) is struggling with the problem of processing the daily volume of foreign-language information efficiently and promptly. Due to internal regulations and security concerns, the use of commercially available, AI-supported translation programs such as ChatGPT & Co. is ruled out, writes “Bild”. The reason: the servers and operators of many such services are located abroad, meaning that the spy agency would risk leaking classified information.

As a result, the foreign intelligence service continues to rely heavily on human translators, Bild claims to have learned. The agency's so-called language service employs a three-digit number of people, including many freelancers on a fee basis. Their task is the “legally compliant” translation of the collected foreign-language files. However, this process is time-consuming; it can take several weeks to process long documents.

Videos by heise

According to the report, the sheer volume of information that the BND collects daily from various sources, such as electronic telecommunications reconnaissance from interception stations like the one in Bad Aibling, as well as inquiries from agents and personal sources, overwhelms the translators' capacities. A particularly critical step in this process is the “pre-assessment,” which is used to assess the relevance of the content. According to high-ranking BND employees, the inadequate preliminary assessment of materials without precise knowledge of the full content poses the risk of important information and strategic objectives being overlooked.

According to a controversial Bundestag resolution, the spies are currently allowed to search 30 percent of internet communication worldwide using selectors. Despite the thousands of search terms used, the results are often poor, but according to the newspaper, hundreds of conversations, emails, chat messages, and text messages get caught in the data vacuum cleaner used by the BND every day. This involves unprovoked mass surveillance in the form of “strategic telecommunications reconnaissance” (“Sigint”).

As the use of commercial AI translation programs is taboo, the BND relies on its own technological solutions. The agency has been developing CAT (Computer Assisted Translation) tools based on in-house developments or collaborations with German companies for two decades. However, according to the report, these programs are used more for content orientation than for a complete translation. The BND is therefore forced to rigorously prioritize and sift out information. It is not known whether the agency cooperates with the much-praised local translation service DeepL. Data sovereignty is also likely to be the sticking point here.

The ongoing importance of human expertise is underlined by a current job advertisement at the BND. Under the reference number AS-2025-300, the service is looking for “freelance translators (m/f/d) on a fee basis,” in particular with knowledge of Eastern European, Asian, or African languages. The authority expressly emphasizes “excellent listening comprehension,” which underlines the relevance of the manual processing of transcripts of conversations.

(nie)

Don't miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.