Deke Kassabian
           Internet2 SALSA Group
and the University of Pennsylvania

                October 14, 2008
    Some security tools require users to install an “agent,” a small
     software component that runs on the users endpoint.
    Examples include:
     ◦    Some asset management tools 
     ◦    Some tools that look for personally identifiable information (PII)
     ◦    Some virus protection tools
     ◦    Some network access control tools 
    Agent software runs on a laptop, desktop or handheld with the
     privileges of the primary user, or with full administrator privileges.
    The agent coordinates with some central console software. 
    This talk explores security and privacy concerns related to
     deployment of such agent-based tools.
    Organizations are encouraged to consider the scope and
     implications of deployment very carefully.
    When considering agent-based security solutions, it is
     important to understand their capabilities.
    Ironically, those software agents with the greatest
     capabilities could raise the greatest concerns. 
    In cases where the agent has very extensive access to
     user data, there could be a significant risk to both
     privacy and security.
  Is the real goal about testing system configuration,
   or about system contents? 
  The difference may be important as you consider
   technical approach and suitability of tools.
  Configuration vulnerabilities can probably be
   found in less intrusive ways than most techniques
   used to audit system contents. 
  If it really is about content, then there may be
   some questions about data administration.
  Who is responsible for the data involved?
  Some organizations have technical staff with a
   “data owner” or “data administrator” role.
  These people see to the responsible handling of
   certain datasets, often from a security and limited
   distribution standpoint.
  Do those who deploy the agent software and
   those responsible for sensitive data sets have the
   same goals and clearance.
    A corporate IT setting may have a different community.
    Users/employees may be predominantly using company-supplied
     computing equipment.
    Common ground: Protection of local networks and other local IT
     assets. 
    But, colleges and universities may have different relationships with
     their user communities than in many corporate IT environments. 
     ◦  Users may be more “transient”
     ◦  Users may be using their computing equipment
     ◦  Users may have different expectations about privacy
    Many of our universities and colleges have a more diverse
     community of users with a range of relationships to the organization.
  Non-faculty employees of the university
  Typically have a traditional employment
   arrangement, including use of employer-supplied
   computing equipment. 
  This community may be the simplest to consider
   for the purposes of this discussion.
  Their use of an employer provided computer is
   likely governed by the terms of their employment,
   and agent-based tools may be very appropriate.
    Members of the faculty may have computing equipment
     acquired in a wide variety of ways.
    Some may have computers supplied by the university.
    Some may be using their own personally purchased
     computers, or computers that are part of some other
     professional relationship they have with another employer. 
    Others may have computers purchased using grant dollars.
    In some of these cases, the faculty member may be
     reluctant to install software whose value they question and
     which may in fact threaten to compromise their privacy or
     agreements with other entities who may have provided the
     computing equipment in question. 
  Like faculty, these users may have computing
   equipment acquired in a wide variety of ways.
  Few have computers supplied by the university.
  They may be more than reluctant to install
   software whose value they question – they may be
   prohibited by the terms of use of the supplier of
   their computer. 
  They may also be unable given the level of
   permissions with which they operate on the
   computer.
    Resident students often arrive on campus with
     personally owned computers. 
     ◦  They use campus networks full time, at all hours of the
        day.
     ◦  In some ways, they are like a home user and we are like
        their ISP.
  They may be reluctant to install software whose
   value they question and which could threaten to
   compromise privacy.
  They may or may not be able to meet explicit
   goals of end user agreements such as appropriate
   use without these tools.
  May have personally owned computers, or laptops
   supplied by employers.
  In the latter case, they may be prohibited from
   installing agents by their employerʼs terms of use.
  In fact, they may be unable given the level of
   permissions they have on the computer.
  These users may have the “weakest” relationship
   with our colleges and universities. 
  They are briefly on campus, and have what they
   believe to be reasonable expectations for network
   access. 
  Do we ask them to install agents for the period of
   their visit? 
  Do we have a way to assure them that they are
   fully removed when they leave our campus?
    Recently, I looked at a few Network Access Control (NAC)
     solutions. 
    One established vendor in this space boasted about the
     capabilities of their agent software.
    Their NAC system could query agents on laptops and
     make access decisions based on fine detail about
     applications and data files present on the end-station,
     about processes currently running, and more.
    As a quick example, they showed an easy to create rule
     that allowed detection of a Windows computer running
     “Notepad”. 
    They went on to describe their extensive capabilities in
     recognizing p2p file sharing tools installed on end-stations.
    At what point do these capabilities become “spyware?”
    If we do deploy “spyware” on endstations on our
     network, do we risk being asked to use these
     extensive capabilities to search for
     ◦  Particular applications, or media files whose names
        suggest material that may be copyright protected?
     ◦  Data that might indicate plagiarism? 
    Will some ask for these capabilities to be used in
     large “sweeps?” For example, do the words
     “plastic” and “explosive” appear close together in
     any files on any user computers?
  Aside from the privacy risks, there are associated
   security risks.
  Throughout the Internet, attacks are increasingly
   being aimed at security products. 
  Each such deployed tool can be an attractive
   target.
  In a case of “too much of a good thing,”
   proliferation of such tools may increase potential
   exposure rather than reducing it. 
  A basic review of some of the available agent-
   based solutions coming into common use shows
   that many are powerful tools that provide
   important information to IT staff, but could as
   easily provide a wealth of valuable information to
   an attacker.
  In some cases, the contents of all files on the
   computer in question are searched as a part of the
   regular operation of the tool. As just one example,
   such a capability mis-applied could prove very
   valuable to an identity thief.
    Agent-based tools can often provide useful functionality
     that helps to meet security, data protection, and asset
     management goals.
    Decision-makers should carefully consider the potential
     risks to both security and privacy associated with such
     deployments.
    Each tool will have associated benefits and risks, and in all
     cases care should be taken in the application of such tools.
     1.  We believe that not every tool is suitable for every environment.
     2.  In some cases a given tool may be more appropriate for certain
         portions of a user community, less appropriate for others. 
     3.  When simpler and less intrusive techniques are available, they
         should be strongly considered.
    Should it be a goal to use the least intrusive means
     to achieve a necessary goal? If so, are agents the
     least intrusive means, or could passive monitoring get
     us close enough?
    Are agents available for all relevant systems, or just
     for a few "mainstream" operating systems? Will that
     discourage diversity, experimentation and selection of
     the best system (rather than the "best system that we
     support"?)  
    In cases where agents are installed, will users be
     fully informed about what the software does and why
     it does it? Is consent required, and if so, what is to be
     done with those who withhold consent? 
    Are these sorts of systems consistent with contracts that
     have been negotiated with organized labor? Some such
     contracts may limit or forbid workplace monitoring of union
     member's computers.
    What is the nature of the communications between the
     agent and the console or system with which it
     communicates? Could someone hijack that reporting
     channel to gain access to confidential information?
    Can the agent be made to only report in university-
     relevant circumstances? (e.g., only if connected to a
     university network connection, but not when the machine is
     connected via a private cable modem or DSL connection?)
    What other affects will the agent have when operating on
     other networks? Could another network's policy (inadvertently)
     prevent connection? Does the agent allow for multiple
     configurations simultaneously, based on network location (e.g.,
     in the event that both organizations use the same agent with
     different configurations)?
    What if there is a conflict over required agents mandated for
     different roles? For example, one venue (such as a university)
     may mandate a particular agent while another venue (such as
     the student's workplace) may mandate another. What if they
     cannot co-exist?
    If the agent based software discovers an issue, is the
     institution always ready to correct the issue now that you've
     been made aware? If not, there may be risk to going out and
     “looking for trouble.”
    If potential criminal activity is discovered, is evidence
     discovered through this mechanism legally-
     admissible, or would it be considered excludable as
     the product of a warrantless search that also lacks
     probable cause? Would the evidence so-collected
     survive evidentiary chain-of-custody challenges?
    If a vulnerability is discovered in the agents
     themselves, how would it be mitigated? Through a
     campus-wide upgrade or uninstall? If those
     capabilities exist, how can we be sure they won't be
     perverted/misused to install malware or to uninstall the
     agent at a strategic time without notice to central IT?
                   Deke Kassabian
           Internet2 SALSA Group
and the University of Pennsylvania

                October 14, 2008