1. The investigation discloses the following:
2.  
3. Two servers in Germany - in Berlin and Nuremberg - are under surveillance by
4. the NSA.
5.  
6. Merely searching the web for the privacy-enhancing software tools outlined in
7. the XKeyscore rules causes the NSA to mark and track the IP address of the
8. person doing the search. Not only are German privacy software users tracked,
9. but the source code shows that privacy software users worldwide are tracked
10. by the NSA.
11.  
12. Among the NSA's targets is the Tor network funded primarily by the US
13. government to aid democracy advocates in authoritarian states.
14.  
15.  The XKeyscore rules reveal that the NSA tracks all connections to a server
16. that hosts part of an anonymous email service at the MIT Computer Science and
17. Artificial Intelligence Laboratory (CSAIL) in Cambridge, Massachusetts. It
18. also records details about visits to a popular internet journal for Linux
19. operating system users called "the Linux Journal - the Original Magazine of
20. the Linux Community", and calls it an "extremist forum".
21.  
22. Disclosure
23.  
24. Three authors of this investigation have personal and professional ties to
25. the Tor Project, an American company mentioned within the following
26. investigation. Jacob Appelbaum is a paid employee of the Tor Project, Aaron
27. Gibson is a paid contractor for the Tor Project, and Leif Ryge is a volunteer
28. contributor to various Tor-related software projects. Their research in this
29. story is wholly independent from the Tor Project and does not reflect the
30. views of the Tor Project in any way. During the course of the investigation,
31. it was further discovered that an additional computer system run by Jacob
32. Appelbaum for his volunteer work with helping to run part of the Tor network
33. was targeted by the NSA. Moreover, all members of this team are Tor users and
34. appear to be have been targets of the mass surveillance described in the
35. investigation.
36.  
37. It is a small server that looks like any of the other dozens in the same row.
38. It is in a large room devoted to computers and computer storage, just like
39. every other room in this industrial park building on Am Tower Street just
40. outside the city of Nuremberg. That the grey building is surrounded by barbed
41. wire seems to indicate that the servers' provider is working hard to secure
42. their customers' data.
43.  
44. Yet despite these efforts, one of the servers is targeted by the NSA.
45.  
46. The IP address 212.212.245.170 is explicitly specified in the rules of the
47. powerful and invasive spy software program XKeyscore. The code is published
48. here exclusively for the first time.
49.  
50. After a year of NSA revelations based on documents that focus on program
51. names and high-level Powerpoint presentations, NDR and WDR are revealing NSA
52. source code that shows how these programs function and how they are
53. implemented in Germany and around the world.
54.  
55. Months of investigation by the German public television broadcasters NDR and
56. WDR, drawing on exclusive access to top secret NSA source code, interviews
57. with former NSA employees, and the review of secret documents of the German
58. government reveal that not only is the server in Nuremberg under observation
59. by the NSA, but so is virtually anyone who has taken an interest in several
60. well-known privacy software systems.
61.  
62. The NSA program XKeyscore is a collection and analysis tool and "a computer
63. network exploitation system", as described in an NSA presentation. It is one
64. of the agency’s most ambitious programs devoted to gathering "nearly
65. everything a user does on the internet." The source code contains several
66. rules that enable agents using XKeyscore to surveil privacy-conscious
67. internet users around the world. The rules published here are specifically
68. directed at the infrastructure and the users of the Tor Network, the Tails
69. operating system, and other privacy-related software.
70.  
71. Tor, also known as The Onion Router, is a network of several thousand
72. volunteer-operated servers, or nodes, that work in concert to conceal Tor
73. users' IP addresses and thus keep them anonymous while online.
74.  
75. Tails is a privacy-focused GNU/Linux-based operating system that runs
76. entirely from an external storage device such as a USB stick or CD. It comes
77. with Tor and other privacy tools pre-installed and configured, and each time
78. it reboots it automatically wipes everything that is not saved on an
79. encrypted persistent storage medium.
80.  
81. Normally a user's online traffic - such as emails, instant messages,
82. searches, or visits to websites - can be attributed to the IP address
83. assigned to them by their internet service provider. When a user goes online
84. over the Tor Network, their connections are relayed through a number of Tor
85. nodes using another layer of encryption between each server such that the
86. first server cannot see where the last server is located and vice-versa.
87.  
88. Tor is used by private individuals who want to conceal their online activity,
89. human rights activists in oppressive regimes such as China and Iran,
90. journalists who want to protect their sources, and even by the U.S. Drug
91. Enforcement Agency in their efforts to infiltrate criminal groups without
92. revealing their identity. The Tor Project is a non-profit charity based in
93. Massachusetts and is primarily funded by government agencies. Thus it is
94. ironic that the Tor Network has become such a high-priority target in the
95. NSA's worldwide surveillance system.
96.  
97. As revealed by the British newspaper The Guardian, there have been repeated
98. efforts to crack the Tor Network and de-anonymize its users. The top secret
99. presentations published in October last year show that Tor is anathema to the
100. NSA. In one presentation, agents refer to the network as "the king of
101. high-secure, low-latency internet anonymity". Another is titled "Tor Stinks".
102. Despite the snide remarks, the agents admit, "We will never be able to
103. de-anonymize all Tor users all the time".
104.  
105. The former NSA director General Keith Alexander stated that all those
106. communicating with encryption will be regarded as terror suspects and will be
107. monitored and stored as a method of prevention, as quoted by the Frankfurter
108. Allgemeine Zeitung in August last year. The top secret source code published
109. here indicates that the NSA is making a concerted effort to combat any and
110. all anonymous spaces that remain on the internet. Merely visiting
111. privacy-related websites is enough for a user's IP address to be logged into
112. an NSA database.
113.  
114. An examination of the XKeyscore rules published here goes beyond the slide
115. presentation and provides a window into the actual instructions given to NSA
116. computers. The code was deployed recently and former NSA employees and
117. experts are convinced that the same code or similar code is still in use
118. today. The XKeyscore rules include elements known as "appids",
119. "fingerprints", and "microplugins".  Each connection a user makes online - to
120. a search engine, for example - can be assigned a single appid and any number
121. of fingerprints.
122.  
123. Appids are unique identifiers for a connection in XKeyscore. Appid rules have
124. weights assigned to them.  When multiple appids match a given connection, the
125. one with the highest weight is chosen. Microplugins may contain software
126. written in general-purpose programming languages, such as C++, which can
127. extract and store specific types of data. The rules specifically target the
128. Tor Project's email and web infrastructure, as well as servers operated by
129. key volunteers in Germany, the United States, Sweden, Austria, and the
130. Netherlands. Beyond being ethically questionable, the attacks on Tor also
131. raise legal concerns.  The IP addresses of Tor servers in the United States
132. are amongst the targets, which could violate the fourth amendment of the US
133. constitution.
134.  
135. The German attorney Thomas Stadler, who specializes in IT law, commented:
136. "The fact that a German citizen is specifically traced by the NSA, in my
137. opinion, justifies the reasonable suspicion of the NSA carrying out secret
138. service activities in Germany. For this reason, the German Federal Public
139. Prosecutor should look into this matter and initiate preliminary
140. proceedings."
141.  
142. One of NSA's German targets is 212.212.245.170.  The string of numbers is an
143. IP address assigned to Sebastian Hahn, a computer science student at the
144. University of Erlangen. Hahn operates the server out of a grey high-security
145. building a few kilometers from where he lives. Hahn, 28 years old and
146. sporting a red beard, volunteers for the Tor Project in his free time. He is
147. especially trusted by the Tor community, as his server is not just a node, it
148. is a so-called Directory Authority. There are nine of these worldwide, and
149. they are central to the Tor Network, as they contain an index of all Tor
150. nodes. A user's traffic is automatically directed to one of the directory
151. authorities to download the newest list of Tor relays generated each hour.
152.  
153. Quellcode NSA  "anonymizer/tor/node/authority" fingerprint.
154.  
155. Hahn's predecessor named the server Gabelmoo, or Fork Man, the nickname of a
156. local statue of Poseidon. After a look at the NSA source code, Hahn quickly
157. found his server's name listed in the XKeyscore rules. "Yes, I recognize the
158. IP address of my Tor server called 'gabelmoo'." he said. "Millions of people
159. use it to stay safe online, and by watching the server and collecting
160. metadata about its users, those people are put at risk." The rule shown to
161. Hahn, published below, has a fingerprint called
162. 'anonymizer/tor/node/authority'. The fingerprint targets users who connect to
163. Gabelmoo and other Tor Directory Authority servers. In Germany, the Tor
164. Directory Authorities like Gabelmoo that are specifically targeted by
165. XKeyscore rules are in Berlin and Nuremberg. Additional targets are located
166. in Austria, Sweden, the United States, and the Netherlands.
167.  
168. Quellcode NSA  Fragments of XKeyscore rules targetting Tor directory
169. authorities.
170.  
171. The expression below performs essentially the same function, but it specifies
172. the Tor directory authorities located in Five Eyes countries (Australia,
173. Canada, New Zealand, the United Kingdom and the United States) separately
174. from those in other countries. As the comment explains, the "goal is to find
175. potential Tor clients connecting to the Tor directory servers."
176.  
177. Another rule catalogs users connecting to known Tor relays. This is not
178. difficult, because the addresses of all normal Tor relays are published by
179. the directory authorities so that the Tor software on users' computers can
180. select its own path through the network. In addition to the public relays,
181. connections characterized as Tor based on protocol identifiers are also
182. cataloged.
183.  
184. Not only Metadata
185.  
186. Internet service providers in countries with strong censorship such as China
187. and Iran frequently block connections to known Tor relays. To avoid this
188. blocking, The Tor Project maintains a list of non-public relays called
189. "bridges" to allow users to avoid this type of blocking. Bridges are run by
190. volunteers and they share the details with the Tor Project to help censored
191. users reach the internet.
192.  
193. Quellcode NSA  Microplugin which extracts bridge addresses from the full text
194. of Tor Project emails.
195.  
196. Users can request a bridge address via email or on the web. The following
197. fingerprints show two ways that XKeyscore attempts to track Tor bridge users.
198. First, the fingerprint "anonymizer/tor/bridge/tls" records connections to the
199. bridges.torproject.org server. Second, in order obtain the actual bridge
200. addresses for the purpose of tracking connections to them in the future, the
201. "microplugin" fingerprint called "anonymizer/tor/bridge/email" extracts data
202. from the body of the emails that the Tor Project sends to its users.
203.  
204. This code demonstrates the ease with which an XKeyscore rule can analyze the
205. full content of intercepted connections. The fingerprint first checks every
206. message using the "email_address" function to see if the message is to or
207. from "bridges@torproject.org". Next, if the address matched, it uses the
208. "email_body" function to search the full content of the email for a
209. particular piece of text - in this case, "https://bridges.torproject.org/".
210. If the "email_body" function finds what it is looking for, it passes the full
211. email text to a C++ program which extracts the bridge addresses and stores
212. them in a database.
213.  
214. Quellcode NSA  Fingerprint to identify visitors to the Tor Project website.
215.  
216. The full content of the email must already be intercepted before this code
217. can analyze it. XKeyscore also keeps track of people who are not using Tor,
218. but who are merely visiting The Tor Project's website (www.torproject.org),
219. as this rule demonstrates:
220.  
221. Quellcode NSA  Rules targeting people viewing the Tails or Linux Journal
222. websites, or performing Tails-related web searches.
223.  
224. It is interesting to note that this rule specifically avoids fingerprinting
225. users believed to be located in Five Eyes countries, while other rules make
226. no such distinction. For instance, the following fingerprint targets users
227. visiting the Tails and Linux Journal websites, or performing certain web
228. searches related to Tails, and makes no distinction about the country of the
229. user.
230.  
231. The comment in the  source code above describes Tails as "a comsec mechanism
232. advocated by extremists on extremist forums". In actuality, the software is
233. used by journalists, human rights activists, and hundreds of thousands of
234. ordinary people who merely wish to protect their privacy.
235.  
236. The rules related to Tails clearly demonstrate how easily web searches and
237. website visits can be spied on by XKeyscore. On June 25, 2014, the United
238. States Supreme Court noted how sensitive this type of information is in their
239. Riley v. California decision against warrantless searches of mobile phones:
240. "An Internet search and browsing history [...] could reveal an individual’s
241. private interests or concerns - perhaps a search for certain symptoms of
242. disease, coupled with frequent visits to WebMD."
243.  
244. Quellcode NSA  C++ program which searches "raw traffic" for .onion addresses.
245.  
246. In addition to anonymous internet access, Tor also provides a mechanism for
247. hosting anonymous internet services called "Hidden Services". These sites'
248. URLs contain a domain name in the pseudo-top-level-domain ".onion" which is
249. only accessible using Tor. The code shown below finds and catalogs URLs for
250. these sites which XKeyscore sees in "raw traffic", creating a unique
251. fingerprint for each onion address. Each .onion address found in raw traffic
252. is extracted and stored in an NSA database:
253.  
254. Quellcode NSA  "anonymizer/mailer/mixminion" appid matching all connections
255. to 128.31.0.34.
256.  
257. There are also rules that target users of numerous other privacy-focused
258. internet services, including HotSpotShield, FreeNet, Centurian,
259. FreeProxies.org, MegaProxy, privacy.li and an anonymous email service called
260. MixMinion as well as its predecessor MixMaster. The appid rule for MixMinion
261. is extremely broad as it matches all traffic to or from the IP address
262. 128.31.0.34, a server located on the MIT campus.
263.  
264. That server is operated by the Tor Project's leader Roger Dingledine, an MIT
265. alumnus. The machine at this IP address provides many services besides
266. MixMinion, and it is also one of the above-mentioned Tor directory
267. authorities. Dingledine said "That computer hosts many websites, ranging from
268. open source gaming libraries to the Privacy Enhancing Technologies Symposium
269. website."
270.  
271. Sebastian Hahn, the Tor volunteer who runs Gabelmoo, was stunned to learn
272. that his hobby could interest the NSA: "This shows that Tor is working well
273. enough that Tor has become a target for the intelligence services. For me
274. this means that I will definitely go ahead with the project.”
275.  
276. When asked for a reaction to the findings, the Tor Project's Roger Dingledine
277. stated the following: "We've been thinking of state surveillance for years
278. because of our work in places where journalists are threatened. Tor's
279. anonymity is based on distributed trust, so observing traffic at one place in
280. the Tor network, even a directory authority, isn't enough to break it. Tor
281. has gone mainstream in the past few years, and its wide diversity of users -
282. from civic-minded individuals and ordinary consumers to activists, law
283. enforcement, and companies - is part of its security. Just learning that
284. somebody visited the Tor or Tails website doesn't tell you whether that
285. person is a journalist source, someone concerned that her Internet Service
286. Provider will learn about her health conditions, or just someone irked that
287. cat videos are blocked in her location. Trying to make a list of Tor's
288. millions of daily users certainly counts as wide scale collection. Their
289. attack on the bridge address distribution service shows their "collect all
290. the things" mentality - it's worth emphasizing that we designed bridges for
291. users in countries like China and Iran, and here we are finding out about
292. attacks by our own country. Does reading the contents of those mails violate
293. the wiretap act? Now I understand how the Google engineers felt when they
294. learned about the attacks on their infrastructure.”
295.  
296. NDR and WDR wanted to know from the NSA how it justified attacking a service
297. funded by the U.S. government, under what legal authority Tor Network users
298. are monitored, and whether the German government has any knowledge of the
299. targeting of servers in Germany. Instead of adressing the questions
300. repeatedly posed to them, the NSA provided the following statement: "In
301. carrying out its mission, NSA collects only what it is authorized by law to
302. collect for valid foreign intelligence purposes - regardless of the technical
303. means used by foreign intelligence targets. The communications of people who
304. are not foreign intelligence targets are of no use to the agency. In January,
305. President Obama issued U.S. Presidential Policy Directive 28, which affirms
306. that all persons - regardless of nationality - have legitimate privacy
307. interests in the handling of their personal information, and that privacy and
308. civil liberties shall be integral considerations in the planning of U.S.
309. signals intelligence activities. The president's  directive also makes clear
310. that the United States does not collect signals intelligence for the purpose
311. of suppressing or burdening criticism or dissent, or for disadvantaging
312. persons based on their ethnicity, race, gender, sexual orientation, or
313. religion. XKeyscore is an analytic tool that is used as a part of NSA's
314. lawful foreign signals intelligence collection system. Such tools have
315. stringent oversight and compliance mechanisms built in at several levels. The
316. use of XKeyscore allows the agency to help defend the nation and protect U.S.
317. and allied troops abroad. All of NSA's operations are conducted in strict
318. accordance with the rule of law, including the President's new directive."
319.  
320. However, the research contradicts the United States' promise to Germany that
321. German citizens are not surveiled without suspicion. Using Tor in Germany
322. does not justify targeting someone, the German attorney Thomas Stadler
323. states: "Tor users do not breach any laws, it is absolutely legitimate to act
324. anonymously on the internet. There are many good reasons to remain
325. anonymous."
326.  
327. What is deep packet inspection?
328.  
329. Deep Packet Inspection, or DPI, refers to the class of technology which
330. examines the content of data packets as they travel across a network. A
331. packet is the fundamental unit of transfer in packet switched networks like
332. the internet. While DPI is commonly used by organizations to monitor their
333. own networks, its use on public networks for censorship and surveillance has
334. been widely condemned by privacy advocates and the United States government
335. alike.
336.  
337. In 2012, the head of the U.S. Delegation to the World Conference on
338. International Telecommunications, U.S. Ambassador Terry Kramer, said “some
339. companies have used deep packet inspection technologies to not look at
340. aggregate customer information, traffic information, et cetera, but to look
341. at individual customer information. So looking at individuals and what sites
342. they’re on and how much capacity they’re using, et cetera, as you can
343. imagine, we’re very much opposed to that because we feel that’s a violation
344. of people’s privacy and gets into, obviously, censorship, et cetera”.
345.  
346. Despite its public political condemnations of invasive DPI use, the United
347. States "Intelligence Community" and its "Five Eyes" partners (Australia,
348. Canada, New Zealand, and the United Kingdom) operate massive internet-scale
349. DPI systems themselves, including XKeyscore. The use of XKeyscore is not
350. limited to these partners, however. The software has been shared with the
351. German BND and BfV, as well as the Swedish FRA, amongst others.
352.  
353. Active vs Passive
354.  
355. XKeyscore and the systems that feed it are considered "passive", meaning that
356. they silently listen but do not transmit anything on the networks that they
357. are targeting. However, through a process known as "tipping", data from these
358. programs can trigger other systems which perform "active" attacks.
359.  
360. Quantum is a family of such programs, including Quantuminsert, Quantumhand,
361. Quantumtheory, Quantumbot, and Quantumcopper, which are used for offensive
362. computer intrusion. Turmoil, Quantum, and other components of the Turbulence
363. architecture are running at so-called "defensive sites" including the
364. Ramstein Air Force base in Germany, Yokota Air Force base in Japan, and
365. numerous military and non-military locations within the United States.
366.  
367. Both Turmoil and XKeyscore feed selected data to real-time "tipping"
368. programs, such as Trafficthief, which can both alert NSA analysts when their
369. targets are communicating and trigger other software programs. Selected data
370. is "promoted" from the local XKeyscore data store to the NSA's so-called
371. "corporate repositories" for long term storage, analysis and exploitation.
372.  
373. More information about XKeyscore
374.  
375. In 2013, the British newspaper The Guardian revealed that by 2008 more than
376. 150 internet surveillance facilities around the world were running the
377. XKeyscore Deep Packet Inspection software. All of the internet traffic
378. observed by XKeyscore, both metadata and full content, is analyzed and stored
379. temporarily at the collection sites for periods ranging from days to weeks,
380. while selected data is forwarded on to other locations for long-term storage.
381.  
382. The storage, indexing, and querying functions are performed at or near the
383. collection sites because the volume of data being collected is too large to
384. forward everything back to facilities in other countries. Analysts working
385. from various locations around the world may search specific XKeyscore sites,
386. or send their queries to a collection of sites.
387.  
388. XKeyscore provides a modular architecture in which tens of thousands of small
389. computer programs, or rules, written in XKeyscore's specialized programming
390. languages called Genesis and XKScript as well as general-purpose languages
391. such as C++ and Python, are run against all traffic to categorize it and
392. extract data. This indexing of the "full take" allows analysts to query the
393. temporary storage stored at the XKeyscore site, effectively sifting through
394. already pilfered communications which occurred before they had deemed them
395. interesting for a specific reason.
396.  
397. XKeyscore can be fed by several different programs, including Wealthycluster
398. and Turmoil. These programs "sessionize" the data, which means that
399. individual connections, such as a request for a web page, are reconstructed
400. from the stream of intercepted packets.
401.  
402. Locations where the NSA runs XKeyscore include Special Source Operations
403. (SSO) sites, typically found at or near major telecommunication providers'
404. infrastructure; Special Collection Service (SCS) sites, usually located
405. inside diplomatic facilities like embassies and consulates; and FORNSAT sites
406. where satellite communications are intercepted. All of these types of sites
407. are known to exist in Germany.
408.  
409. Other "Five Eyes" partners also operate XKeyscore installations. The United
410. Kingdom's Tempora program runs the largest instance of XKeyscore. Both the
411. software itself and limited access to NSA databases have been shared with
412. so-called "3rd party" partners including Germany. The German foreign
413. intelligence agency BND and the domestic intelligence agency BfV are testing
414. the Software.