3. Who is nickg?
Web Application background
Software Development background
Linux/Unix background
Most everything was either social media and/or
ecommerce since 1994
I started at Etsy two years ago. There was no one
dedicated on fraud and security in engineering.
A lot of this was learned the hard way
My perspective on fraud is probably a bit different.
Season to taste.
4. Who is Etsy?
“Online marketplace for creative small businesses”
No inventory, marketplace. Instead we have both sides
– Buyer risk
– Seller risk
When fraud happens, it‟s not silent. It‟s public.
We lose trust (and money).
We are very sensitive to fraud and risk == a lot of R&D
5. What and Where is Risk?
Many types of risk… but today we‟ll talk about
Fraud
Security
Internal Threats
Business Continuity
Physical Security
Intellectual property
6. Thinking about Risk and Fraud
“System working correctly, but with stolen or false
credentials causing financial loss”
Constant, always happening.
More business focused
Continuous output (“fraud is 1%”)
Think: stolen credit cards, bogus seller that doesn’t ship
goods.
7. Thinking about Risk and Application Security
“System working incorrectly when given invalid or unexpected
input, causing financial loss, data loss/theft, system downtime,
vandalism, or attack on another system.”
Unexploited problems exists, now.
Can be costly dealing with compliance, disclosure, legal.
More technical-focused
Binary Output (“we are breached, or not”)
Think: SQLi, XSS, buffer overflow attacks, data breach, etc
Of course, security flaws can be used to commit fraud
8. Account Takeover Blurs the Line
Account takeover crosses the boundaries from
site security to personal member security.
Problems can be public
Fraud and Security two sides of same coin.
14. Log It
Leverage existing centralized logging (if not get it)
You can index it – lots of 3rd party solutions
Make new security/fraud/sensitive data log or
namespace
Log this:
– Password resets
– Email changes
– Credit card changes
– Login failures
Also great for internal risk monitoring.. Who is doing what
15. Graph It
Critical for visibility and promotion or your pain points
TechOps is likely using Ganglia and/or Graphite
Enhance the application using gmetric and/or StatsD
Example: Login Success and Failures.
16. Monitor It
Now that you are logging and graphing, can you monitor
and alert on outliers?
Likely Nagios or another system in place
Don’t worry, Etsy is ok. This was from a dead machine.
17. PSA #1: Start the dialog for 100% SSL
SSL isn‟t just for login and checkout
Entire categories of risk are eliminated with 100%
Little to no additional load on infrastructure.
Evaluate your current setup at Qualsys SSL Labs
https://www.ssllabs.com/
Get an “A” with Apache/OpenSSL using*
SSLProtocol -all +TLSv1 +SSLv3 SSLCipherSuite
HIGH:MEDIUM:!aNULL:+SHA1:+MD5:+HIGH:+MEDIUM
(*) Assuming your patches are up-to-date
19. Using the QA infrastructure
Zooming out, QA / Fraud / Security begin to look the
same
A serious bug might be indistinguishable from fraud
QA typically tests positive flows
Fraud Engineering leverages QA to test negative flows.
http://jenkins-ci.org/
20. Test Your Invariants
Things that should be always true (or false).
Super easy to test
– “This page should always be SSL”
– “This page should always require login”
– “http://..../server-status” doesn‟t display to public”
– “http://…/wp-admin” requires a password”
– “This page should never displays the full credit card”
– “Google never visits this page”
You‟ll be amazed or frightened by the results
22. Use the central log to find…
Syntax errors from the database!
Certainly a bug, but perhaps SQLi attempts
Uncaught DatabaseException: 42601 7 ERROR:
syntax error at position 2 near "&" in
SELECT COUNT(*) FROM convos
WHERE uid = ? AND names LIKE „?‟
with [895724897,"Ll'or1=1"]
at DBConnection.php
based on a true story
all queries and values changed to protect the guilty
23. … or find this?
Ungraceful exits
Really should never happen
Latent bug? Need to upgrade? Or probing attack?
24. Even if you can’t fix it,
establish the base line and
look for deviations from it
26. Product should be helping with
The delicate balance between
easy enough so you don‟t loose customers
vs. hard enough so attackers go elsewhere
vs. the barriers appropriate to risk.
27. Can you make security a desired feature?
Can you offer your best customers better security
solutions so they don‟t have account takeover?
Has anyone even asked them?
Not necessarily resulting in more engineering work.
– Site messaging improvements
– Outreach
– Customer education
How can you make account takeover recovery easier?
How can you message the user when they their email got
erased or if they ?
28. BizOps
Have you talked to the email marketing and/or online-ad
targeting groups?
The work they do is oddly similar to fraud analytics.
– Breakdown by sales by country over time
– Customer visit frequency by sales
– Average purchase price
– Basket Analysis
Helping them make their data more real time/visible
helps the business and adds additional eyes on fraud
30. Fraud Engineering
There is certainly pure fraud engineering:
– Integration with risk management solutions
– Rule and model building
– Analysis and reporting
– Behavior tracking
And there is certainly security engineering
– Authentication and Authorization
– CSRF / SQLi protections
– Secure coding initiatives https://buildsecurityin.us-cert.gov/
But there is a lot more you can leverage from the
organization.
31. Work on preventing false positives
Eliminating false positives helps your risk management
system work better.
Disable form submit buttons after being pressed
(prevents double clicks)
Add rate limits to just about everything on the site
Does not necessarily stop determined attackers, but…
if someone is breaking or bumping up against your rate
limits, you know they are up to something.
32. PSA #2: No passwords in plain text!
I beg of you.
Also don‟t store them as plain MD5 or SHA1
Use a “salted hash” system.
Start the process today!
33. Here’s a secret
Your engineers are bored.
90% of a computer science degree isn‟t used on a day to
day basis
This is why open source projects exists: to work on cool
stuff they can‟t do at work.
They have side-projects already
There is a huge cognitive surplus is sitting around.
34. Here’s another
This laptop is the equivalent of at least 8 Amazon EC2
“small” instances and has a terabyte of storage.
“Hard problems” such as machine learning, natural
language processing, big data are rapidly being
commoditized.
There is a huge computational surplus laying around the
office.
35. Now that you know the secret, use it
Fraud problems are engineer-bait -- it‟s full of fun hard
problems
Leverage your employees! Advertise your problems.
If that fails, find interns! I‟m sure your local schools will
be happy to help.
37. Customer Service
They know more than you on how the site is working and
performing.
All fraud ends up being a customer service problem
Improving customer service == improving fraud
management.
Talk to them and build the best #(&^$*# tools that you
can for them.
Gains of 4x-5x can occur by eliminating crap out of their
workflow.
39. Case Study
Customer Service was looking into some “problematic
customers.” Login history didn‟t really make much sense.
Got bounced to fraud engineering.
40. Case Study
Looking into the IP addresses, and doing whois showed
many were coming from “rent-a-slice” datacenters.
Linode, Amazon, and Rackspace are used as an example. They are great companies and are
recommend. No implication of wrong doing should be implied!
41. Case Study
This lead to a side-project mapping the range of IP
addresses that belong to rent-a-slice centers.
43. Case Study
Product is ok with throwing up CAPTCHAs on these
accounts in certain cases since it‟s unlikely to interfere with
the vast majority of users.
http://www.google.com/recaptcha
44. Case Study
Customer Service tool updated so reps can see if IP is a
datacenter or not, and have direct access to whois
Note: no implication that the hosting provider is or has done anything wrong.
They might be victims of fraud themselves.
45. Case Study
Oddly many users are legit (privacy nuts? escaping great
firewall of china?)
Working on CS/Product strategy to reach out to the legit
customers on why.
Rolling out analysis to checkout/purchase.
Would love your feedback and help, so….
46. Case Study: Our List is Yours
Over 25,000,000 total IP addresses
Over 1700 IP blocks
Over 350 providers
https://github.com/client9/ipcat