Why is Internet Security So Hard? Dr. Stephen Kent Chief Scientist- Information Security.

Why is Internet Security So Hard?

Dr. Stephen Kent

Chief Scientist- Information Security

Internet Security Security for the Internet includes both security for

network operations and security for network users The former is usually the purview of ISPs, the

latter is a shared responsibility among users, ISPs, and vendors

For network users, there is a need to secure information on computers and in transit across the Internet

This presentation focuses on security for Internet users

What is Security?

ISO 7498-2 defines five security services Confidentiality (secrecy) Authentication (identify verification) Integrity Access control Non-repudiation (not “taking back” what one “said”)

Users also would likely include Preventing spam Preventing denial of service Privacy …

Information Security Disciplines Physical security

Procedural security

Personnel security

Compromising emanations security

Operating system security

Communications security

a failure in any of these areas can undermine the security of a system

Security Terminology

Vulnerabilities security flaws in systems

Attacks means of exploiting vulnerabilities

Countermeasures technical or procedural means of addressing

vulnerabilities or thwarting specific attacks

Threats motivated adversaries capable of mounting attacks

which exploit vulnerabilities

Adversaries (The Bad Guys)

Hackers Disgruntled employees Industrial spies Terrorists Special interest groups Journalists Real spies Criminals (organized or otherwise)

Adversary Characteristics Capabilities

Network wiretapping Remote attacks against operating systems or

applications “Social engineering” (e.g., SPAM) Physical attacks Personnel subversion

Resources Personnel Technology Funds

Aversion to detection

Vulnerabilities The simple characterization of our problem is the

existence of vulnerabilities in products We face a two sorts of vulnerability problems:

Known vulnerabilities Unknown vulnerabilities

For known vulnerabilities we can deploy specific countermeasures

For unknown vulnerabilities, at best we try to prevent/detect behavior that might be exploiting these vulnerabilities

Sources of Vulnerabilities

Design flaws operating system & application vulnerabilities protocol design vulnerabilities

Implementation flaws programming errors undocumented system & application “features”

Mismanagement unintended and/or residual authorizations failure to deploy security bug fixes

Security Continuum

There are no perfect, secure systems Systems are "adequately secure" only relative to a

perceived threat Absence of obvious insecurities is not a good

indication that a system is adequately secure Risk analysis, if properly performed, provides a

methodology for identifying what constitutes adequate security

The Threshold Effect

Once a technical attack against a security technology has been "debugged" it can be executed by a wide range of (inexperienced) attackers

A technical attack that can be effected using inexpensive hardware or software is especially easy to transfer from sophisticated attackers to amateurs

Thus it is dangerous to dismiss an attack as "too complex or too technical" because the perceived attackers do not possess the technical capability to mount the attack

Why are the Bad Guys Winning?

Most vendor software has poor security characteristics Too complex Badly designed Buggy

Most users are sloppy Don’t install the latest patches Easily tricked (social engineering) Poor password choices, password reuse, …

Hackers value their time at 0, but user have other priorities in life!

Common Defense Strategies

Firewalls Intrusion Detection Systems Anti-virus technology (in hosts and in mail

gateways) Anti-spam technology (in hosts and in mail

gateways) Periodic penetration testing (enterprise nets) Centralized patch management (enterprise nets) Anti-DOS mechanisms (ISPs)

Firewalls Recently renamed Intrusion Prevention Devices

(IPDs), probably to help sell more of them :-) The term covers a wide range of technologies,

from simple, stateless packet filtering, to application-specific devices

At the low end, these offer minimal protection against most adversaries

At the high end they are expensive and often interact badly with new applications

In all cases, management of the firewall rule sets is complex, time consuming, and thus imperfect

Intrusion Detection Systems (IDSs) An IDS attempts to:

Detect behavior that exploits known vulnerabilities Detect behavior that might exploit some class of unknown

vulnerabilities Detect behavior that might be a precursor to an attack

IDS may attempt to: Detect signatures of known attacks Detect anomalous behavior Do both

IDS’s tend to work poorly, because of the ambiguities associated with attempts to deal with unknown attacks or to define “normal” behavior

False positives (incorrect flagging of traffic as “evil” is common, and makes these systems hard to use

Anti-virus Systems

These attempt to detect viruses (and worms), typically distributed via e-mail attachments or other forms of file transfer

Usually they are signature based, which means they know only about previously-detected viruses

A network manager or user has to acquire signature list updates periodically, or become vulnerable to newer viruses

These can be effective if properly managed, but people are sloppy, and virus writers are prolific

Anti-spam Technology The problem with spam is that it is impossible to

distinguish from legitimate mail, in the worst case Some anti-spam technology works on signatures,

like anti-virus technology, but it is not very effective because spam generation software does not focus on software vulnerabilities, like viruses

Some anti-spam technology is based on Baysean filters (probabilistic measures), but it too is subject to false positive/false negative tuning problems

Spam is of value to its senders primarily because users are greedy or naïve; solving this is NOT a technical problem

Penetration Testing

This is an approach used by many enterprises, but rarely by individual users

At the low end it is automated, mostly a patch check on end systems and a firewall filer rules check

At the high end one pays “experts” to try to break into your system(s)

The low end is useful as a form of external checking re good housekeeping

The high end is very expensive

Centralized Patch Management The notion here is to enable an IT organization to

check the status of end systems and to patch them before the systems are successfully attacked

Vendors like Cisco and Microsoft offer this as a service, part of “admission control” to a LAN

This is another form of “good housekeeping” checking, on a more frequent basis

It is analogous to low end penetration testing, a form of centrally managed anti-virus updating

BUT, an already-compromised system can avoid detection if the attacker is clever

Anti-DoS Technology Denial of service attacks seek to make resources

unavailable, typically through overloading network access lines with lots of traffic

The problem is that it is hard to tell good traffic from bad traffic out in the Internet (vs. at an end system)

Some systems try to look at traffic flows and discard packets if the flow to a given destination is “too high” BUT, good traffic is often discarded as well as bad!

We know that some DoS hackers have thousands of “zombie” systems available to them, dispersed over the Internet, to launch attacks, which makes it almost impossible to counter such attacks without causing problems for legitimate users as well

Abstraction and Attacks

One strategy for an adversary is to attack below the layer of abstraction at which security measures are defined, or via ill-defined interfaces

Complex applications and operating systems like Windows have many ill-defined interfaces

Security measures implemented in applications (or middleware) embody high levels of abstraction

The trend is to create more opportunities for an attacker as we use more complex, high level application development environments, e.g., web services

Security in Products: Functionality vs. Assurance Security functions: usually visible, security-

relevant features that provide the means by which security is invoked and managed

Security assurance: often invisible means by which one develops confidence in the correct operation of security features

Many products now advertise lots of security functions (because today, security sells), but the products offer little or no assurance!

Security Assurance

Product security assurance techniques penetration testing detailed code review use of formal specifications security evaluation criteria

Unfortunately, these techniques are either very expensive or very haphazard

As a result, we have few products for which we have a good idea of their security quality

Security & Privacy: A Quick Look

Security Accountability Uniform identification Extensive auditing Correlation of audit data Centralized management Mediated access to all

data

Privacy Anonymity Use of diverse identifiers Limited data collection No sharing of records Distributed autonomy Mediated access to

records that affect privacy

Security and privacy need not be in conflict, but it takes a lot of effort to balance the two

Conclusions Internet security is hard because:

Its hard to counter unknown vulnerabilities in products Even security products themselves often have unknown

vulnerabilities The utility of an IDS is limited by feature rich

environments Most CIOs can’t even track all the systems in their nets There is no methodology for designing a secure system

from secure components (and we have few secure components anyway)

Abstraction favors the adversary Some problems (e.g., spam) are not technical in nature People are sloppy, greedy, and sometimes naïve

Questions