Download Design for Values - (DIMACS) Rutgers

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Document related concepts

Social perception wikipedia, lookup

Trust (emotion) wikipedia, lookup

Transcript
Phishing - A Social Disorder
L. Jean Camp
Alla Genkina
Allan Friedman
www.ljean.com
A tangent
Workshop on the Economics of Information Security
2- 3 June 2005
Reception June 1
Early registration rates end tomorrow
http://www.infosecon.net/workshop
ROI in security investment, economics of spam, economics of identity,
vulnerability markets, empirical investigation on privacy & security
investments
Phishing - A Social Disorder

A lack of context

Human trust behaviors

Social engineering
– Begin with the social problem not with the technical potential
– Solve for human trust behaviors
– Provide unique signals
Human and Computer Trust
Social
sciences
Experiments designed to evaluate how people extend trust
Game theory
Common assumption: information exposure == trust
Philosophy
Macro approach
Examine societies and cultural practices
Computer
Security
Build devices to enable trust
Philosophy Suggests

Trust is necessary to simplify life
¯ People have an innate desire or need to trust
¯ People will default to extending trust

People will continue to extend trust - so create
another source of trust don’t defeat trusting
behaviors
Research on Humans Suggest...

Humans may not differentiate between machines

Humans become more trusting of ‘the network’

Humans begin with too much trust
Confirmed by philosophical macro observation
Confirmed by computer security incidents
Validated
• E-mail based
Scams
Viruses
Hoaxes
Three Observations

Humans respond differently to human or computer
"betrayals" in terms of forgiveness?

People interacting with a computer do not distinguish
between computers as individuals but rather respond to their
experience with "computers”

The tendency to differentiate between remote machines
increases with computer experience
Response to Failure

Humans respond differently to human or computer
"betrayals" in terms of forgiveness
– Attacks which are viewed as failures as ‘ignored’ or forgiven
– Technical failures as seen as accidents rather than design
decisions
» May explain why people tolerate repeated security failures
– May inform the balance between false positives and negatives in
intrusion detection
» Rarely identified malicious behavior taken more seriously
» Technical failures easily forgiven
» Failures expected
Individiation

People interacting with a computer do not
distinguish between computers as individuals but
rather respond to their experience with
"computers”
– People become more trusting
– People differentiate less
– People learn to trust
» First observed by Sproull in computer scientists in 1991
» Confirmed by all future privacy experiments
Differentiation

The tendency to differentiate between remote
machines decreases with computer experience
– Explicit implication of second hypothesis
– Explains common logon/passwords
» along with cognitive limits
Observed Verification of
Hypotheses

Users are bad security managers
 PGP, P3P,….

Security should necessarily be a default

Does end-to-end security maximize autonomy without end-toend human abilities and tendencies?

Data currently being compiled on experiments

Surveys illustrate a continuing confusion of privacy &
security
Computer security is built for
machines

Passwords
 Humans are a bad source of entropy

SSL




Two categories: secure and not secure
Does not encourage differentiation
Every site should include a unique graphic with the lock
Computer security should seek to differentiate machines
Privacy standards are built for
machines

P3P assumes
– All merchants trustworthy w.r.t. their own policies
– Assumes increasingly sophisticated user
– One standard for all transactions

PGP
– Monotonic increase in trust
– No reset
– No decrease in rate of trust extension
» To compensate for increasing trust
– No global or local reset
» E.g. change in status
Key revocation is built for
Machines

CRL tend to be single level

Different levels of revocation are needed
– Falsified initial credential
» All past transactions suspect
– Change in status
» Future transactions prohibited
– Refusal of renewal
» Current systems adequate

CRL should reflect the entire systems in which they work,
including the social system

CRL is too simplistic, depends on active checking
WHAT TO DO?

Computers
–
–
–
–

Process data
Store data
Transmit data
Distinguish
» atomicity, privacy, availability,
Humans
– Understand context
– Evaluate uncertainty
– Make lumping decisions based on context

Begin with the human as the basis of the design
– Examine human interactions
– Signal humans using pre-existing social capital
Context

Trust is contextual

Phillips on Zero Knowledge
– Nyms had to be selected before the person engaged in
interaction
– The interaction in question is entering information
– The information should be available before the interaction
Not Even Communicating with
Users

Identity theft
– unauthorized use of authenticating information to assert identity in the
financial namespace
– Internal process violation - Choicepoint (at least 145k records)
» All access to the Choiepoint database was authorized
» Subsequent misuse was authorized by the information obtained via
Choicepoint
– Security Violation - Berkeley
– Confidentiality information - Bank of American backup data 1.2M
records

Risk profile is similar for individuals in all three cases
Dominant Trust Communication
Equivalent Value
Cradle to Grave ID…. So What

Authentication as what? For what?

Identification as having what attributes?

Scope of namespace
– License to drive
» requires driving test
– SSN
» taxpayer ID to assert right to benefits
– Birth certificate
» proof of age
– Define a credit namespace that allows for large scale illegal employment
– Require that credit and banking agencies handle their own risks and
pay for their own identity schemes for all transactions but cash
– Make video rental agencies accept their own risks
– Cell phone requires that you have an account to pay for it
– DL requires you know how to drive
Perfect Single ID
… for every namespace
… and every context
… for all people
for definitions: http:// www.ljean.com/
… or solve the problem at hand by
enabling contextually rational trust
behavior
Embedding Browsing in Social
Context

First trust challenge
– Enabling trust to allow entry onto the net
– Enabling monetary flows

Second trust challenge
– Providing meaning trust information
» TrustE, BBB, Verisign
– Namespaces for specific trust assertions
» Christian, gay friendly, responsible merchants
– Requires a common understanding of the limits of the namespaces
» Transitivity
» Automated trust decisions
» Consistency across contexts or explicit definition of context
• E.g., purchase a book
– On divorce
– On impotency
– On effective job searching
– On number theory
Enabling Trust Behavior

Signal not to trust

Combine trust perceptions for end users
– Privacy
» Based on personal experience
» Or verification of centralized authority (BBB)
– Reliability
» Personal experience
» Verification (Consumer reports)
– Security
» Is personal experience valuable here?
• Q: what is the value of peer production for security information
» Centralized verification (Counterpane)
– ID theft vs. account misuse is distinguished by the bank but not by the
customer
– Loss of data from privacy or security is the same for the individual
» For whom should we design
Context

Selected context
determines

Social network
display

Shared
information

NOT certificate
authorities

Depends on
homophily
Visual Trust

Verisign will protect you from anyone who will not give
them money
– There has been no business plan of a single trusted root which
aligns that root with the end user.

There are competitive trust sources that align with user

Uses pop-up blocker
Centralized Elements

No hierarchies
– Trust is not inherently transitive
– “Verisign is the trust company”

Certificates
– Signed green list

Signer determines
–
–
–
–
Frequency of update
Business model
Determinant of entry
Potential examples
» FDIC
» Consumer reports
» BBB
» Phishguard
Reputation

Initial reputation of zero

First visit goes to 1 (out of ten)

After second visit it increases

Each visit decreases rate of delay
– Max of 10

Explicit rating
– Stays constant without alteration over time
A New Paradigm for Design

Design technology to conform to user behaviors

Assume individuals will default to trust, then lump, and
forgive

Depends upon
»
»
»
»
»

Usability
Reputation system design
Homophily
Storage capacity
Efficient search
Provide signals NOT to trust
– Do not assume that no signal means no trust.
– No signal will be interpreted as trust
Definitions

Attribute. a characteristic associated with an entity, such as an individual. Examples of persistent attributes
include height, eye color, and date of birth. Examples of temporary attributes include address, employer,
organizational role. A Social Security Number are an example of a long-lived attribute.
and
Some biometrics data
are persistent, some change over time or can be changed, (e.g., fingerprints and hair color).
Personal identifier.
persistent identifiers associated with an individual human
and the attributes that are
difficult or impossible to alter. For example, human date of birth, height, and genetic pattern.
Anonym (as in anonymous). An identifier associated with no personal identifier, but only with a single-use
attestation of an attribute. An anonymous identifier identifies an attribute, once. An anonymous identifier
used more than once becomes a pseudonym.
Pseudonym. An identifier associated with attributes or sets of transactions, but with no permanent identifier
More Definitions

Identification.
Association of a personal identifier with an individual presenting attributes. For example,
accepting the association between a physical person and claimed name; or determining an association with a
medical record and a patient using physical attributes.
Authentication. Proving as association between
an identifier or attribute. For example, the association of an
automobile with a license plate or a person with an account. The automobile is identified by the license plate,
it is authenticated as legitimate by the database of cars that are not being sought for enforcement purposes.
Identity Authentication. Proving as association between an entity and an identifier. For example, the
association of
a person with a credit or educational record.
Attribute Authentication. Proving as association between an entity and an attribute. For example, the
association of an painting with a certificate of authenticity.
association between entity and identifier
established.
This
is usually a two step process, where the
is established, and then the link to identifier and attribute is
Yielding

Authorization. A decision to allow a particular action based on an identifier or attribute. Examples include
the ability of a person to make claims on lines of credit; the right of an emergency vehicle to pass through a
red light; or a certification of a
Identity. That set of
radiation-hardened device to be attached to a satellite under construction.
permanent or long-lived temporal attributes associated with an entity