{ "title": "XKEYSCORE Tech Strings in Documents (aka Tech Extractor) 2009", "type": "REPORTED", "description": "A December 2009 TOP SECRET XKeyscore training presentation on the Tech Extractor (Tech Strings in Documents) — an early context-aware scanning capability that tags keywords only when they appear inside email, chat and document bodies rather than anywhere in raw internet traffic, letting analysts find technically interesting documents by terms too generic for traditional content selection, with an example flagging a Pakistani document mentioning WiMAX and DVB satellite technology by scanning inside the file.", "long_description": "This is a TOP SECRET//COMINT//REL TO USA, AUS, CAN, GBR, NZL XKeyscore training presentation dated December 2009 on the Tech Extractor, also called Tech Strings in Documents. It describes XKeyscore's first implementation of context-aware scanning — a refinement of content-based selection designed to reduce the volume of junk results that traditional \"soft selection\" produced.\n\nThe presentation frames the problem with existing selection engines: they were originally designed to handle TELEX traffic, a highly formatted, content-rich message type that bears little resemblance to raw DNI internet traffic containing HTML, web pages and base-64 encoded documents. When an analyst tasks a Boolean query like \"bomb\" AND \"chemical,\" they want communications mentioning those terms, not every web page, news story or blog post where the words happen to appear. The Tech Extractor addresses this by making selection context-aware — knowing where inside the raw DNI a term appears. The presentation uses a Crocodile Dundee reference (\"That's not a knife... THAT's a knife\") to contrast blunt soft selection with the \"surgical\" precision of contextual dictionaries.\n\nThe Tech Extractor focuses on three contexts: email bodies, chat bodies and document bodies (Microsoft Word, Excel, PowerPoint, Project and Visio; Adobe PDF and PostScript; and Rich Text Format). It works by scanning a list of keywords against those three contexts and tagging matching sessions with metadata — explicitly not filtering, selecting or forwarding data home, but tagging sessions much as AppIDs and fingerprints do. Analysts can then use these tags as part of USSID-18 compliant queries, though the presentation notes that Tech Extractor tags are not USSID-18 compliant by themselves and may need to be combined with a valid foreign IP, MAC address or country code. The term lists come from analysts, who supply the XKeyscore team with \"Tech Dictionaries\" containing multiple category names (\"Tech Names\").\n\nThe worked example demonstrates the capability's value. A Pakistani document (\"NIB Ranchor Line KHI.doc\") was tagged with three hits: the tech name \"wireless\" for the value WIMAX, \"satellite\" for DVB, and \"mac\" for a MAC address (00200E00953C) found inside the document. The presentation notes this document would have been nearly impossible to find with traditional methods — the terms \"wimax\" and \"dvb\" are too generic for CADENCE-style tasking, and the MAC address hit relied on an anchorless regular expression that corporate scanning engines could not perform. A second example shows a Nokia phone repair email (IMEI, model 6300) sent by a service centre coded GCC-KHI, and a third demonstrates full foreign-language support by tagging common Arabic expressions in emails from Pakistan's tribal regions.\n\nThe document is significant as the origin point for XKeyscore's context-aware scanning, which later evolved into the more sophisticated context-sensitive fingerprinting documented in the AppID presentations (html_title, http_host, email_body, doc_body and similar contexts). It demonstrates the analytical rationale behind context-aware selection — distinguishing communications content from incidental web content — and shows how the technique enabled discovery of technically valuable documents (network equipment specifications, WiMAX and satellite deployment details, MAC addresses) that traditional keyword selection would have buried in noise. The Pakistani tribal-region focus of the examples situates the capability within counterterrorism collection.", "tags": [ "context-aware scanning", "soft selection", "content-based selection", "email body", "chat body", "document body", "Microsoft Office", "TELEX", "DNI", "Nokia", "IMEI", "Karachi", "tribal regions", "metadata tagging", "target discovery" ], "document": { "creator": "NSA", "date_created": "2009-12", "classification": [ "TOP SECRET", "COMINT" ], "dissemination": [ "REL TO USA, AUS, CAN, GBR, NZL" ], "banner": "TOP SECRET//COMINT//REL TO USA, AUS, CAN, GBR, NZL", "genre": "slides", "programs": [ "XKEYSCORE", "CADENCE", "UIS" ], "countries": [ "Pakistan" ], "people": [] }, "publication": [ { "publisher": "The Intercept", "date_published": "2015-07-01", "title": "XKEYSCORE: NSA’s Google for the World’s Private Communications", "link": "https://theintercept.com/2015/07/01/nsas-google-worlds-private-communications/", "reporters": [ "Marquis-Boire, Morgan", "Greenwald, Glenn", "Lee, Micah" ], "contributors": [], "doc_source": "https://theintercept.com/document/2015/07/01/xks-tech-extractor-2009/", "doc_url": "https://s3.documentcloud.org/documents/2116496/xks-tech-extractor-2009.pdf" } ], "aka_id": [], "pdf_sha256": "5479741c928afc690ede67e52ab520a76899bc53c08a95e9fd6c66ea38c752f8", "pages": 14 }