diff options
| author | Owen Jacobson <owen.jacobson@grimoire.ca> | 2013-10-29 21:02:20 -0400 |
|---|---|---|
| committer | Owen Jacobson <owen.jacobson@grimoire.ca> | 2013-10-29 21:02:20 -0400 |
| commit | 86204c92eb7f1e32f799fb14748aba159721a0d8 (patch) | |
| tree | 8165aedee3017c675087657d601dd5fb14185595 | |
| parent | 7dcff9999d1f46ce28f1b356643378f99b7edaf1 (diff) | |
Fuck Intro.
| -rw-r--r-- | wiki/ethics/linkedin-intro.md | 181 |
1 files changed, 181 insertions, 0 deletions
diff --git a/wiki/ethics/linkedin-intro.md b/wiki/ethics/linkedin-intro.md new file mode 100644 index 0000000..3944786 --- /dev/null +++ b/wiki/ethics/linkedin-intro.md @@ -0,0 +1,181 @@ +# LinkedIn Intro is Unethical Software + +[LinkedIn Intro](https://intro.linkedin.com) is a mail filtering service +provided by LinkedIn that inserts LinkedIn relationship data into the user's +incoming and outgoing mail. This allows, for example, LinkedIn to decorate +incoming mail with a toolbar linking to the sender's LinkedIn account, and +automatically injects a short "signature" of your LinkedIn profile into +outgoing mail. + +These are useful features, and the resulting interaction is quite smooth. +However, the implementation has deep, unsolvable ethical problems. + +LinkedIn Intro reconfigures the user's mobile device, replacing their mail +accounts with proxy mail accounts that use LinkedIn's incoming and outgoing +mail servers. All of LinkedIn's user-facing features are implemented using +HTML and JavaScript injected directly into the email message. + +## Password Concerns + +LinkedIn Intro's proxy mail server must be able to log into the user's real +incoming mail server to retrieve mail, and often must log into the user's real +outgoing mail server to deliver mail with correct SPF or DKIM validation. This +implies that LinkedIn Intro must know the user's email credentials, which it +acquires from their mobile device. Since this is a "use" of a password, not +merely a "validation" of an incoming password, the password must be available +_to LinkedIn_ as plain text*. There are two serious problems with this that +are directly LinkedIn's responsibilty, and a third that's indirect but +important. (Some email providers - notably Google - support non-password, +revokable authentication mechanisms for exactly this sort of use. It's not +clear whether LinkedIn Intro uses these safer mechanisms, but it doesn't +materially change my point.) + +LinkedIn has a somewhat unhappy security history. In 2012, they had a +[security +breach](http://www.nytimes.com/2012/06/11/technology/linkedin-breach-exposes-light-security-even-at-data-companies.html) +that exposed part of their authentication database to the internet. While they +have very likely tightened up safeguards in response, it's unclear whether +those include a cultural change towards more secure practices. Certainly, it +will take longer than the year that's passed for them to build better trust +from the technical community. + +Worse, the breach revealed that LinkedIn was actively disregarding known +problems with password storage for authentication. [Since at least the late +70's](http://cm.bell-labs.com/cm/cs/who/dmr/passwd.ps), the security community +has been broadly aware of weaknesses of unsalted hash-based password +obfuscation. More recently, [it's become +clear](http://www.win.tue.nl/cccc/sha-1-challenge.html) that CPU-optimized +hash algorithms (including MD5 and both SHA-1 and SHA-2) are weak protection +against massively parallel password cracking — cracking that's quite cheap +using modern GPUs. Algorithms like +[bcrypt](http://codahale.com/how-to-safely-store-a-password/) which address +this specific weakness have been available since the late 90's. LinkedIn's +leaked password database was stored using unsalted SHA-1 digests, suggesting +either a lack of research or a lack of understanding of the security +implications of their password system. + +Rebuilding trust after this kind of public shaming should have involved a +major, visible shift in the company's culture. There's easy marketing among +techies — a major portion of LinkedIn's audience, even now — to be done by +showing how on the ball you can be about protecting their data; none of this +marketing has appeared. The impact of raising the priority of security issues +throughout product development should be visible from the outside, as risky +features get pushed aside to address more fundamental security issues; no such +shift in priorities has been visible. It is reasonable, observing LinkedIn's +behaviour in the last year, to conclude that LinkedIn, as a company, still +treats data security as an easy problem to be solved with as little effort as +possible. This is not a good basis on which to ask users to hand over their +email passwords. + +While the security community has been making real efforts to educate users to +use a unique password for each service they use, the sad reality is that most +users still use the same password for everything. As LinkedIn Intro must +necessarily store _plain text_ passwords, it will be a very attractive target +for future break-ins, for employee malfeasance, and for United States court +orders. + +## What Gets Seen + +LinkedIn Intro is not selective. Every email that passes through an +Intro-enabled email account is visible, entirely, to LinkedIn. The fact that +the email occurred is fodder for their recommendation engine and for any other +analysis they care to run. The contents may be retained indefinitely, outside +of either the sender's or the recipients' control. LinkedIn is in a position +to claim that Intro users have given it _permission_ to be intrusive into +their email in this way. + +Very few people use a dedicated email account for "corporate networking" and +recruiting activities. A CEO (LinkedIn's own example) recieves mail pertaining +to many sensitive aspects of a corporation's running: lawsuit notices, gossip +among the exec team, planning emails discussing the future of the company, +financials, email related to external partnerships at the C*O level, and many, +many other things. LinkedIn's real userbase, recruiters and work-seeking +people, often use the same email account for LinkedIn and for unrelated +private activities. LinkedIn _has no business_ reading these emails or even +knowing of their existence, but Intro provides no way to restrict what +LinkedIn sees. + +Users in heavily-regulated industries, such as health care or finance, may be +exposing their whole organization to government interventions by using Intro, +as LinkedIn is not known to be HIPAA, SOX, or PCI compliant. + +The resulting "who mailed what to whom" database is hugely valuable. I expect +LinkedIn to be banking on this; such a corpus of conversational data would +greatly help them develop new features targetting specific groups of users, +and could improve the overall effectiveness of their recommendation engine. +However, it's also valuable to others; as above, this information would be a +gold mine for marketers, a target for break-ins, and, worryingly, _immensely_ +useful to the United States' intelligence apparatus (who can obtain court +orders preventing LinkedIn from discussing their requests, to boot). + +(LinkedIn's recommendation engine also has issues; it's notorious for +[recommending people to their own +ex-partners](http://community.linkedin.com/questions/31650/linkedin-sent-an-ex-girlfriend-a-request-to-someon.html) +and to people actively suing one another. Giving it more data to work with +makes this more likely, especially when the data is largely unrelated to +professional concerns..) + +LinkedIn Intro's injected HTML is also suspect by default. Tracking email open +rates is standard practice for email marketing, but Intro allows _LinkedIn_ to +track the open rate of emails _you send_ and of emails _you recieve_, +regardless of whether those emails pertain to LinkedIn's primary business or +not. + +## User Education + +All of the risks outlined above are manageable. With proper information, the +end user can make an informed decision as to whether + +* to ignore Intro at all, or +* to use Intro with a dedicated "LinkedIn Only" email account, or +* to use Intro with everything + +LinkedIn's own marketing materials outline _absolutely none_ of these risks. +They're designed, as most app landing materials are, to make the path to +downloading and configuring Intro as smooth and unthreatening as possible: the +option to install the application is presented before the page describes what +the app _does_, and it never describes how the app _works_ — that information +is never stated outright, not even in Intro's own +[FAQ](https://intro.linkedin.com/micro/faq). Witholding the risks from users +vastly increases the chances of a user making a decision they aren't +comfortable with, or that increases their own risk of social or legal problems +down the road. + +## LinkedIn's Response + +Shortly after Intro's first round of public mockery, a LinkedIn employee +[posted a +resonse](http://blog.linkedin.com/2013/10/26/the-facts-about-linkedin-intro/) +to some of the security concerns. The post is interesting, and I recommend you +read it. + +The key point about the response is that it underscores how secure Intro is +_for LinkedIn_. It does absolutely nothign to discuss how LinkedIn is curating +its users' security needs. In particular: + +> We isolated Intro in a separate network segment and implemented a +> tight security perimeter across trust boundaries. + +A breach in LinkedIn proper may not imply a breach in LinkedIn Intro, and vice +versa, but there must be at least some data passing back and forth for Intro +to operate. The nature and structure of the security mechanisms that permit +the "right" kind of data are not elaborated on; it's impossible to decide how +well they actually insulate Intro from LinkedIn. Furthermore, a breach in +LinkedIn Intro is still incredibly damaging even if it doesn't span LinkedIn +itself. + +> Our internal team of experienced testers also penetration-tested the +> final implementation, and we worked closely with the Intro team to +> make sure identified vulnerabilities were addressed. + +This doesn't address the serious concerns with LinkedIn Intro's _intended_ +use; it also doesn't do much to help users understand how thorough the testing +was or to understand who vetted the results. + +## The Bottom Line + +The software industry is young, and immature, and wealthy. There is no ethics +body to complain to; had the developers of Intro said "no", they would very +likely have been replaced by another round of developers who would help +LinkedIn violate their users' privacy. That does not excuse LinkedIn; their +product is vile, and must not be tolerated in the market. |
