Content Acquisition Optimization PowerPoint Slides

Open PDF
Creating agency: NSA National Security Agency (US)
Document date: -
Pages: 9
Genre: slides
Banner:
Classification: TOP SECRET SI
Dissemination: NOFORN
Part of an assembly: SSO Collection Optimization

An NSA presentation on content acquisition optimisation — describing how incidental bulk collection from Yahoo Webmessenger, email address books, buddy lists and webmail inboxes was generating massive unintended collection volumes.

Long Description:
This is a TOP SECRET//SI//NOFORN NSA presentation on content acquisition optimisation, addressing the problem of incidental bulk collection generated by SSO's major access programmes when targeted selectors appear in address books, buddy lists and webmail inboxes of non-targeted contacts. The Yahoo Webmessenger section describes sporadic collection of 30,000-60,000 sessions per day from update data sent to users logged into Yahoo's Instant Messenger, with intermittent bursts of 20,000 or more sessions against contacts of a single targeted selector — none of which are the target themselves and none of which carry owner attribution. Over a dozen selectors were detasked in two weeks because a target's contact was idling on Yahoo Webmessenger, including selectors described as timely for the Libyan transition and Greek financial crisis. The address books section shows that at a snapshot on 10 January 2012, approximately 22% of all traffic across SSO's major accesses was address book collection — up from 12% in August 2011. The table shows five SIGADs: US-3171, DS-200B, US-3261, US-3145 and US-3180, with a combined total of 712,366 address books out of 3,257,738 total sessions. Of 689,246 sessions collected, only 95,086 — 13.8% — were attributable to known users. Facebook collection showed the highest attribution rate at 95.87%, while Hotmail showed the lowest at 1.06%. The document notes address books are stored redundantly across MARINA, MAINWAY, PINWALE and CLOUDS. The buddy lists and inboxes section notes that on a representative day NSA collected approximately 500,000 buddy lists and inboxes, with more than 90% collected solely because a tasked selector appeared as a contact — not as a communicant, content owner or active user. The case study illustrates how these dynamics compounded. An Iran Quds Force-related Yahoo account had its address book hacked and began sending spam to its contact list in September 2011. Because the contact list included Yahoo groups with hundreds or thousands of members, collection at DS-200B spiked dramatically — the spam messages themselves were collected, then inboxes of recipients were collected each time they were viewed, then address books were collected each time recipients who had added the hacked account opened their own address books. Collection spiked to tens of thousands of megabytes per day before the selector was emergency detasked on 20 October 2011. The presentation estimates that memorialising to PINWALE only address books and inboxes owned by target selectors would have reduced PINWALE volumes by over 90%. The Mobile IMAP section briefly notes that Android's IMAP email implementation uses excessive bandwidth due to the protocol not being designed for devices with intermittent connections, contributing additional unintended collection volume.
Programs:
MARINA, MAINWAY, PINWALE, CLOUDS, XKEYSCORE, MUSCULAR, SSO

People mentioned:
-

Countries:
Libya, Greece, Iran

Tags:
emails, IMAP, buddy lists, overcollection, mobile phones, webmail, Android, Yahoo, Yahoo Messenger, metadata, Hotmail, Gmail, Facebook, address books, incidental collection, content acquisition, spam, detasking, bulk collection

Identifier:
ACLU_-_a-sso-content-optimization-detailed-redacted
Also known as:
EFF_-_20131014-wapo-content_acquisition_optimization2

PDF SHA-256:
ca3a440cc785f518f23f99106b95878da5a8c3a3287d1560c552e3c97474322e

TOP SECRET//SI//NOFORN

SPECIAL SOURCE OPERATIONS

Special Source Operations

Content Acquisition Optimization

[REDACTED]
[REDACTED]
[REDACTED]

TOP SECRET//SI//NOFORN


TOP SECRET//SI//NOFORN

Yahoo Webmessenger

  • Update data sent to individuals logged into Yahoo's Instant Messenger service online
    • Online contact status, unread emails in Yahoo inbox
    • Usually small sessions (2-4kB)
  • Sporadic collection (30,000 – 60,000 sessions per day)
  • Intermittent bursts of collection against contacts of targets
    • Large numbers of sessions (20,000+) against a single targeted selector
    • Not collected against the target (online presence/unread email from target)
    • No owner attribution (metadata value limited to fact-of comms for emails, online presence events for buddies)
  • Over a dozen selectors detasked in two weeks
    • Because a target's contact was using/idling on Yahoo Webmessenger
    • Several very timely selectors (Libyan transition, Greek financial related)

TOP SECRET//SI//NOFORN


TOP SECRET//SI//NOFORN

Address Books

  • Email address books for most major webmail are collected as stand-alone sessions (no content present*)
  • Address books are repetitive, large, and metadata-rich
  • Data is stored multiple times (MARINA/MAINWAY, PINWALE, CLOUDs)
  • Fewer and fewer address books attributable to users, targets
  • Address books account for ~ 22% of SSO's major accesses (up from ~ 12% in August)
Access (10 Jan 12) Total Sessions Address Books
US-3171 1488453 237067 (16% of traffic)
DS-200B 938378 311113 (33% of traffic)
US-3261 94132 2477 (3% of traffic)
US-3145 177663 29336 (16% of traffic)
US-3180 269794 40409 (15% of traffic)
US-3180 (16 Dec 11) 289318 91964 (32% of traffic)
TOTAL 3257738 712366 (22% of traffic)
Provider Collected Attributed Attributed%
Yahoo 444743 11009 2.48%
Hotmail 105068 1115 1.06%
Gmail 33697 2350 6.97%
Facebook 82857 79437 95.87%
Other 22881 1175 5.14%
TOTAL 689246 95086 13.80%

TOP SECRET//SI//NOFORN


TOP SECRET//SI//NOFORN

Buddy Lists,Inboxes

  • Unlike address books, frequently contain content data
    • Offline messages, buddy icon updates, other data included
    • Webmail inboxes increasingly include email content
    • Most collection is due to the presence of a target on a buddy list where the communication is not to, from, or about that target
  • NSA collects, on a representative day, ~ 500,000 buddylists and inboxes
    • More than 90% collected because tasked selectors identified only as contacts (not communicant, content, or owner)
  • Identifying buddylists and inboxes without content (or without useful content) an ongoing challenge

TOP SECRET//SI//NOFORN


TOP SECRET//SI//NOFORN

Scenario: [REDACTED]@yahoo

  • [REDACTED] Sep 2011 [REDACTED]@yahoo.com (tasked S2E, asw Iran Quds Force) has his/her Yahoo account hacked by an unknown actor, sends out spam email to his/her contact list:
DNI Parser Webmail Display YAHOO! MAIL
Active user: [REDACTED]

Message

Date 2011-09-[REDACTED]
Subject [REDACTED]!!!!! (New)
From [REDACTED]@yahoo.com>
To [REDACTED]@yahoogroups.com

[REDACTED]

TOP SECRET//SI//NOFORN


TOP SECRET//SI//NOFORN

Scenario: [REDACTED]@yahoo

  • [REDACTED]@yahoo.com has a number of Yahoo groups in his/her contact list, some with many hundreds or thousands of members
  • At DS-200B in particular, collection spiked as:
    • The initial spam messages were sent (and collected)
    • Inboxes of email recipients were viewed by [REDACTED] contact list
    • Messages were sometimes viewed, but more often sent as precached views on Google and Yahoo (along with inboxes)
    • Inboxes where the recipient did not delete the spam message continued to be collected every time they were viewed
    • Some recipients added [REDACTED]@yahoo.com to their address books (possibly as a spam defeat?) – address books were collected every time

TOP SECRET//SI//NOFORN


TOP SECRET//SI//NOFORN

Scenario: [REDACTED]@yahoo

DS-200B Collection By Day - 11 Sep - 24 Sep (in MB)

120000
100000
80000
60000
40000
20000
0
9/11/2011 9/12/2011 9/13/2011 9/14/2011 9/15/2011 9/16/2011 9/17/2011 9/18/2011 9/19/2011 9/20/2011 9/21/2011 9/22/2011 9/23/2011 9/24/2011

DS-200B Collection By Hour – 18 Sep – 23 Sep (in MB)

40000
30000
20000
10000
0
9/18/2011 0:00 9/18/2011 12:00 9/19/2011 8:00 9/19/2011 20:00 9/20/2011 8:00 9/20/2011 20:00 9/21/2011 8:00 9/21/2011 20:00 9/22/2011 8:00 9/22/2011 20:00 9/23/2011 8:00 9/23/2011 20:00

TOP SECRET//SI//NOFORN


TOP SECRET//SI//NOFORN

Scenario: [REDACTED]@yahoo

  • [REDACTED]@yahoo.com emergency detasked from DS-200B and US-3171 at 13:04Z on 20 Oct
  • Numerous first-order address books and inboxes collected meant tasked selectors on address books or buddy lists of contacts of [REDACTED]@yahoo.com also affected:
    • [REDACTED]@yahoo.com and [REDACTED]@gmail.com emergency detasked off US-3171 at 13:10Z on 20 Sep
  • Memorializing to PINWALE only address books and inboxes owned by target selectors would have reduced PINWALE volumes 90%+
    • Site XKEYSCOREs would buffer data for SIGDEV purposes
    • Metadata from known owner address books and inboxes stored regardless

TOP SECRET//SI//NOFORN


TOP SECRET//SI//NOFORN

Mobile IMAP

  • IMAP protocol used by email clients to fetch mail from server(s)
  • Not designed for devices with intermittent connections (i.e. mobile phones)
  • Android implementation in particular uses a lot of bandwidth
A0 CAPABILITY
A1 LOGIN [REDACTED]
A2 CAPABILITY
A3 EXAMINE INBOX
A4 LIST "" INBOX
A5 LIST "" "INBOX.%"
A6 SEARCH SINCE 15-Aug-2011 UNDELETED ALL
A7 FETCH 17 (ENVELOPE INTERNALDATE RFC822.SIZE
A8 FETCH 17 (BODY.PEEK[HEADER])
A9 CLOSE
A10 LOGOUT
Date From To Subject Attachments
Fri Aug [REDACTED] [REDACTED] [REDACTED] 2nd Payment Reminder [REDACTED] 0
Display Information: Email
  Subject: 2nd Payment Reminder [REDACTED]
  From: [REDACTED]
  To: [REDACTED]
  Date: Fri Aug [REDACTED]

DNI Parser: Document or message has no data

Text Size  View Full Screen

TOP SECRET//SI//NOFORN