4
SentryHQ’s Reactive Security: The New Host-Based Intrusion Detection Paradigm Traditional host intrusion detection systems usually bring an attack to an operator's attention, but this asynchronous attack response paradigm may not be sufficient to stop an attack before it can do damage to a system. The solution, Amr Ali and Zach Dexter explain, is reactive security, or shutting down attacks in real-time, via collaborative attack vector closure. Introduction There are two problems with traditional host intrusion detection systems (HIDS). First, the attack may not be detected, because traditional HIDS may not pick up on sophisticated attacks or attacks that use new vectors. Second, even if an attack is detected, damage may occur before an operator can respond. To solve the first problem, we propose an open-source repository of attack detectors. To solve the second problem, we propose a system to respond to attacks in real-time, including an open-source repository of attack responses. Traditional host intrusion detection systems can abstractly detect attacks that are either identical to previous attacks or similar to previous attacks. The former can be detected using signature databases, and the latter by machine learning algorithms. Pattern recognition is an inductive approach to intrusion detection: The HIDS infers that an attack is taking place when an event shares characteristics of a prior attack. But what if we know nothing about an attack? What if an attack uses a new vector? Traditional host intrusion detection systems solve this problem by letting the attack happen and notify an operator, or partially sever access to the system while being indifferent to the nature of the attack or the operations of the system itself. A traditional HIDS will update a rule database or exclude the just-discovered attack vector from a training set of data on what constitutes the normal behavior of a system. The problem with the traditional HIDS approach is that it seeks to stop poorly-executed or non- coordinated attacks that look like previous attacks or otherwise fail to fool a system. To stop sophisticated or new attacks, we must get the HIDS to deduce that an attack is occurring, even if the HIDS has no knowledge whatsoever of the attack vector, and even if an attack is clever enough to appear to machine-learning algorithms as normal behavior. A deductive security system would specify a set of invariants. If even one of those invariants changes, a system is said to be compromised. But how can any HIDS specify a set of invariants large enough to provide meaningful coverage of attack vectors? How can a deductive HIDS know how invariants might change across deployments to different machines? And what does it mean to close an attack vector that the deductive HIDS doesn't even know is open?

SentryHQ's Reactive Security

  • Upload
    amr-ali

  • View
    190

  • Download
    0

Embed Size (px)

DESCRIPTION

Traditional Host Intrusion detection systems usually bring an attack to an operator's attention, but this asynchronous attack response paradigm may not be sufficient to stop an attack before it can do damage to a system. The solution, Amr Ali and Zach Dexter explain, is reactive security, or shutting down attacks in real-time, via collaborative attack vector closure.

Citation preview

Page 1: SentryHQ's Reactive Security

SentryHQ’s Reactive Security: The New Host-Based Intrusion Detection Paradigm

Traditional host intrusion detection systems usually bring an attack to an operator's attention, but this asynchronous attack response paradigm may not be sufficient to stop an attack before

it can do damage to a system. The solution, Amr Ali and Zach Dexter explain, is reactive security, or shutting down attacks in real-time, via collaborative attack vector closure.

Introduction

There are two problems with traditional host intrusion detection systems (HIDS). First, the attack may not be detected, because traditional HIDS may not pick up on sophisticated attacks or attacks that use new vectors. Second, even if an attack is detected, damage may occur before an operator can respond. To solve the first problem, we propose an open-source repository of attack detectors. To solve the second problem, we propose a system to respond to attacks in real-time, including an open-source repository of attack responses.

Traditional host intrusion detection systems can abstractly detect attacks that are either identical to previous attacks or similar to previous attacks. The former can be detected using signature databases, and the latter by machine learning algorithms.

Pattern recognition is an inductive approach to intrusion detection: The HIDS infers that an attack is taking place when an event shares characteristics of a prior attack. But what if we know nothing about an attack? What if an attack uses a new vector? Traditional host intrusion detection systems solve this problem by letting the attack happen and notify an operator, or partially sever access to the system while being indifferent to the nature of the attack or the operations of the system itself.

A traditional HIDS will update a rule database or exclude the just-discovered attack vector from a training set of data on what constitutes the normal behavior of a system. The problem with the traditional HIDS approach is that it seeks to stop poorly-executed or non-coordinated attacks that look like previous attacks or otherwise fail to fool a system.

To stop sophisticated or new attacks, we must get the HIDS to deduce that an attack is occurring, even if the HIDS has no knowledge whatsoever of the attack vector, and even if an attack is clever enough to appear to machine-learning algorithms as normal behavior.

A deductive security system would specify a set of invariants. If even one of those invariants changes, a system is said to be compromised. But how can any HIDS specify a set of invariants large enough to provide meaningful coverage of attack vectors? How can a deductive HIDS know how invariants might change across deployments to different machines? And what does it mean to close an attack vector that the deductive HIDS doesn't even know is open?

Page 2: SentryHQ's Reactive Security

The answer is collaborative attack vector closure, an easy-to-understand adaptation of open-source software engineering principles. A community of attack detector authors contributes to an open-source repository of invariants.

Let us call an invariant expressed via a programming language a detector. Contributors generalize the invariants so that members of the community may clone any detector, provide parameters relevant to their implementations, and store the customized detectors in a private repository. After testing the customized detectors in the field, community members may commit patches and merge them upstream to the main repository of detectors.

We propose to grow this open-source platform with a good number of invariants for most pieces of software running on today's systems. Attackers will have little opportunity to avoid detection, as the attack vectors that the HIDS doesn't even know about are now closed.

Once an attack is detected, a pre-configured response executes as a countermeasure with the intention to either eliminate the threat or act as means of damage control.

Threat Detection & Response

There are predominantly two methodologies to address threats; either develop detection methods for the specificities of different threats, or detect anomalies in the behavior of the system and treat them as possible threats.

The approach of which a detection method is developed to identify a particular threat works well if and only if we know intrinsic details of the threat we are trying to address. Since the possibilities of threats and their mutations are theoretically infinite, this approach is ultimately a never ending cat-mouse chase. However, this approach also comes with the advantage of facing less false-positives because through it we know how the threat exactly behaves and how we could respond to it.

On the other hand we could compile a set of invariants of a system which we know how it behaves so that we abstractly normalize its operations and be able to detect a threat through sensing anomalies in its behavior. This method comes with the disadvantage of being abstract and thus prone to false-positives due to a legitimate change of an operation or an unforeseeable logical branch in a well behaving set of instructions.

There is no one concrete solution to addressing threats but a combination of variants of methodologies. Since humans are the cornerstone of the reason security as a philosophy exists, we propose that people engage in catering the different security approaches to their unique needs by allowing the community a platform which offers the tools necessary to descriptively define their systems’ operations and at the same time be able to define anticipated threats and how to respond to them.

SentryHQ’s the platform that allows the community to develop detectors for unbounded number of operating systems and applications, detectors that are synchronously combined with responses that are not only catered to the underlying operating system but also to the applications running on-top of it. The locally synchronized combination of a detector and response acts as a first layer of defense against any threat.

Page 3: SentryHQ's Reactive Security

Real-Time Attack Mitigation

We believe that the only way to stop an attack is to prepare an automatic, immediate response before the attack happens. Too often, an attack is over, and the attacker has made off with valuable data, before the attack is noticed. Even if a traditional Host Intrusion Detection System notices the attack, often nothing is done until an operator responds. SentryHQ bolsters the operator’s capabilities by allowing her to configure immediate attack responses.

Attack detectors pick up on abnormal behavior, either by specifying invariant conditions or by looking for signs of an attack. When a detector picks up on an attack, it does more than simply notify the machine’s operator.

The detector fires any number of attack responses on any number of machines. If there is an attack response listening on the compromised machine, the response can shut the attacker out of that machine. If responses on other machines are also listening on the detector, those responses will fire, too. Such flexibility allows the security context to not only be bound to a single machine but an entire network of machines that can realize an attack and respond to it in unanimity.

Collaborative Attack Vector Closure

Work together to achieve maximum attack vector coverage. Community members tag attack detection and attack response code to place it in the public repository. Any member of the community can clone these items, upgrade them, and publish the code back to the public repository. Users can leave code with modifications specific to their machines unpublished.

Over time, SentryHQ will severely restrict the number of vectors still open to attackers. For each detector made available to the community, an attack vector is closed. And each response available in the public repository gives community members more power to stop attacks before they result in damage.

Why SentryHQ?

No one likes nor have the time to dive into endless amount of configuration files and foreign syntaxes to be able to run a HIDS that only reports back a compromise. Beyond initial configuration of your account; SentryHQ components are entirely configurable through our intuitive hosted web interface.

In cyberspace the first “O” of Boyd’s OODA (Observe, Orient, Decide, Act) loop is always impaired; with SentryHQ we’ve managed to enable your entire network to be able to observe an elaborated attack and respond to it on many levels and in dynamic configurations through the detectors and responses deployed on any number of machines you have.

Page 4: SentryHQ's Reactive Security

SentryHQ utilizes a custom implementation of SSHv2 protocol for all of its communication, be it internal or external. We’ve taken great care to account for the worst situations a system can be into including ours.

With SentryHQ you can benefit from and participate in an ever-growing community that constantly supplies the public repository with new detectors and responses that addresses threats that target most applications and their underlying operating systems.