SOC Mistake #5: You treat the SOC as a Project, not a Programme

Building an effective and efficient Security Operations Centre can take a matter of years.  Yes, you can build a foundational level of capability in several months (and it's what companies used to pay me to do in my previous role), but it takes time for processes to be tuned and become muscle memory to the analysts; to filter out false positives and minimise the risk of a real attack sneaking in as a false negative; initial training for your SOC staff is a baseline, understanding business operations, stakeholder priorities and the practical application of their training - these all take time. 

The transition from the foundational to the capability is achieved through repeatable and measureable processes, with the metrics providing the required telemetry for the SOC manager to make operational decisions - "train analyst x in y"; "retire use case z because the operational overhead exceeds it's value to the business"; and "reinforce processes to analyst b who isn't following the required playbooks".

This often means that it can take 2 - 3 years for a SOC to reach it's optimum maturity level (depending on what the business has defined that as) - during this time there will be new threat actors, new vulnerabilities, new systems implemented within the organisation and new defences.  All of these need to be taken into account - putting a stake in the ground today on day one of the SOC build and saying we're going to build this capability, using these event sources with these use cases over a 3 year period is a honourable endeveavour, but by time you finish in 2018, you'll have a SOC fit for 2015.  This is one of the reasons that when I was at HP we encouraged customers to see their requirements as fluid and likely to change overtime, whether this was due to discovering their baseline level of maturity was well off the mark after we started work; that the nature of their infrastructure or business changed or whether budgets got cut - the reality is the end point is a moving target and the best way to get there is small incremental steps.

With this in mind, consideration needs to be taken into account of not just the deployment of a SIEM and log management solution, along with recruiting and training some analysts, but to the ongoing operation of the SOC and the continual investment in time, effort and money needed to keep the SOC on target.

New Company, New Challenges

After four great years at the Security Intelligence & Operations Practice in HP I have moved to a new role this week as the Executive Director of Cyber Risk Management, Corporate & Investment Banking for JP Morgan Chase.  

I'm hoping without the high level of international travel I was doing at HP I will be able to update this blog more frequently - starting with finally finishing the series "Top 10 SOC mistakes!, which has been a long-time coming. 

SOC Mistake #6: You don’t focus on the big picture

This is a mistake we see a lot in Security Operations Centres that have SIEM Use Cases that have been built using a bottom-up approach.  I discussed this in my post SOC Mistake #7: On Use Cases, You Model Your Defences, Not Your Attackers, where SIEM Use Cases are are arrived upon by looking at what event sources are easy to obtain or are available, rather than what is needed to maximise the efficiency and effectiveness of a SIEM’s capability to detect attacks against your key line-of-business infrastructure.

Often in Security Operations Centres that have been built this way, rules cannot be built in a granular enough way to provide the analyst with enough context to determine whether something is a false positive or a real attack without significant digging around.  To deal with event volume the Security Operations Centre has to invest significant cost in hiring additional Level 1 Analysts to perform event triage.  Another problem with bottom-up rules is that they can be extremely tricky to tune, usually have simple correlations relying on two or three different log sources and simple logic – often tuning them for one scenario may detune the detection capability of another.

In contrast a top-down approach should provide multiple opportunities to detect the attack along the attack chain, so if one component of the staged rule is causing mis-fires into the SIEM it is possible to tune anywhere along the staged rules comprising the attack chain.  With this approach you can start to tune out false positives (alarms where there is no real event) without introducing excessive false negatives (missing a real event).  The business impact and threat assessment you undertook as a part of your Use Case Workshop  should drive what the tolerable level of false negatives is: you compare the operational cost of the additional staffing to handle the false positives you have to keep in because you can’t tune them out without introducing false negatives, against the likelihood and impact of the event if you miss it – of course you can’t make these kinds of judgements if you haven’t taken a top-down approach.  I’ll talk more in a later post about collecting and analysing metrics that tell you that you should stop tuning and just can a Use Case and start again with a different approach to it.
Going back to the “Big Picture”, in these bottom-up  Security Operations Centres you often see SIEM events in the triage console that resemble the raw events from the event sources, i.e. they don’t resemble ‘actionable intelligence’.  In fact in one multinational company that paid a significant amount for their SIEM infrastructure we saw the SIEM platform only receiving events from a single device type, from the same manufacturer – they might have well just dunked the Analyst in front of the management console of the device and forgot SIEM altogether.  What constitutes ‘actionable intelligence’ will differ depending on which SIEM-vendors marketing glossies your reading, but to me it is enough information for a Level 1 Analyst to conduct initial triage without having to use a large number of investigatory tools to be able to triage false positives, determine the likely impact of the event on the organisation and determine the level of skill and possible motivations of the attacker.

A Use Case built using the top-down approach will provide this information.  The process of building these kinds of Use Cases involve the modelling of vulnerabilities, threats and controls in the people, processes, applications, data, networks, compute and storage for each line-of-business, armed with the information about where in the attack chain the attack has been detected and all of the event information up to the point of detection (or beyond if the rule also triggered a higher-level of proactive monitoring, such as full packet capture or logging of keystrokes or even redirected to attacker to a tarpit to gather further information on their intent, tools, techniques and procedures).  This information allows the analyst conducting the triage, at a glance, to make an initial determinations around impact, capability and scope of the attack.

The SIEM platform, ideally, should provide integrated tools for further analysis such as the retrieval and visualisation of related historical logs to look for anomalies, correlations, affinity groups and context; as well as the ability to lookup sources IPs, packet captures or executables against threat intelligence sources- and beyond to query the configuration management or identity management servers to understand the use and recent configuration changes to machines, as well as the rights of users, involved.  In fact in HP ArcSight this data can be automatically brought in to enrich the event before it is even opened by the Level 1 Analyst to make them more operationally efficient.

So what is the “Big Picture”, well the answer to that is understanding the Who? What? When? How?, and most difficult, Why? of the attack.  Faced with a huge deluge of rule fires that require significant effort to investigate and of which a large proportion end up being false positives which you “never seem to be able to tune out” when something that looks like a real attack is found, the Analyst often will run around with their hair on fire.  Often they’ll escalate without answering these basic questions and when C-level exec has been got out of bed they’ll ask relevant questions that often the SOC can’t answer – Who? What? When? How? and Why?
Before an incident is declared the function of a Security Operations Centre in large organisation is to answer those questions – to prepare, to detect and to investigate.  They should be able to prioritise the incidents to be dealt with by understanding the capability of the adversary, the impact of the incident and the scope of systems involved.  This is the information they should be passing to the incident responders to allow them to contain, eradicate, finally IT operations works with the SOC to eliminate the vulnerabilities/apply additional controls and then recover (and increasing logging to detect if the machine is attacked again).  During the whole process the Security Operations Centre should be working iteratively with the incident response team and IT operations.

Bad Use Cases that provide no context of the attack, bad integration of intrusion detection tools, lack of knowledge of context of systems and users, coupled with a lack of analytical skills in Analysts results in the focusing on the individual events, not the scope and impact of a potential incident or breach.

One story we frequently tell is of a SOC we knew of that where they hadn’t reached out to IT Operations department to win their buy-in for obtaining logs.  Due to the adversarial relationship with IT Ops and the infosec department the infosec team relied on the logs that they could obtain easily, i.e. the ones from the systems they had ownership of – namely intrusion detection, firewall and anti-virus.  Now everyone who works in information security reading this blog knows just how effective these technologies are in 2015 so nothing was triggering the correlations on the SIEM platform (the customer had also just deployed the default content from the vendor, not tuning it to their available resources).  Funnily enough the SIEM didn’t detect a large breach very public breach that the customer suffered.
Questions were asked by the CEO about why he wasn’t notified and then why the SIEM product they’d spent so much money on had “failed”.  At least as a result of the incident the information security team got carte blanche access to whatever logs they wanted – great right?  Well no.  The small SOC then on-boarded every single log source they could lay their hands on using a bottom-up approach.  The result was chaos – masses of events bleeding into the console providing no answers to the contextual questions and in an overcompensation for not notifying the CEO of the original incident, the SOC team call him out-of-hours over a dozen times in one month over incidents that they had panicked over as they hadn’t been able to truly understanding what was happening.


SOC Mistake #7: On Use Cases, You Model Your Defences, Not Your Attackers

Use Cases - these are simply the most misunderstood subject around both security operations and Security Information & Event Management (SIEM).  

SIEM is one of the most mis-sold and mis-brought items of information security technology.  The most common type of Request For Information we get from customers lists the different types of event sources and the Events Per Second (EPS) for each event source.  The customer often believes that a SIEM is magical, that it understands everything about their business and their threats.  All they need to do is merely pump their raw events into and actionable security intelligence pops out of the other end.  

This means that procurement of SIEM is largely down to whether the SIEM supports the event source and whether the architecture supports the event volume.  This technical bias when purchasing a SIEM means that Request For Information rarely looks at the operational processes in actually using a SIEM, and therefore the effectiveness and efficiency of the SIEM.

This approach ends up aligning with the defensive infrastructure of the organisation, rather than trying to detect the behaviour of an attackers.  The SIEM will typically provide you details of a correlated event or anomaly which is the starting point, not that end point of intrusion analysis - if you SOC staff are simply escalating alerts to the CERT, they are simply acting as 'click monkeys' and not providing any value to the intrusion detection/investigation/response capability.  The more accurate the modelling of the method of attack and additional context the SIEM has around the user, assets and know attacks and attacker behaviour: the more the technology is doing the heavy lifting.  

Ideally what you are looking to achieve is something along the lines of:

  • Level 1: elimination of false positives; prioritisation based on the criticality or sensitivity of the assets involved in the case; initial investigation of the incident to answer the When? Who? How? What? Why? questions before escalation (if required) to the next level for more specialist investigation.
  • Level 2: more detailed investigation of the incident, which may involve escalation to the Incident Response Team (IRT) and then an interactive process of the Level 2 Analyst providing details of compromised systems and suspicious users to the Incident Response Team who will perform more detailed forensic analysis, that may result in one-or-more additional Indicators-of-Compromise (IoC) that the Incident Response Team will pass back to the Level 2 Analyst to conduct searches using the log management platform and SIEM to identify additional systems or users involved in the incident that can then be passed back to the Incident Response Team.  The aim is to allow the SOC to specialise in technical skills required in log analysis and SIEM, and the Incident Response Team in network and host forensics.

When you consider the whole point of a SIEM Use Case is to drive the detection of the attack through correlation or anomaly detection rules in the SIEM platform, to detect incidents through post event analytics and to drive the intrusion analysis process to answer the core questions:

  • When?  When did this attack start?  Is it a stand alone action, or is it a part of a wide campaign or a stage in a broader attack?  If it still ongoing?  Is it better to escalate immediately and kick off the containment and eradication, or to observe with your finger over the trigger to gain more information around intent and capabilities?  With actors that would be typically classified as Advanced Persistent Threats (APTs), rather like the boy with his finger in the dyke, responding to the symptom and closing off one vulnerability will simply mean they'll try again and come back through another avenue.  Gaining a better understanding of the capabilities, tools and ultimate intent of the attacker will help you ensure that your preventative controls will be optimised to prevent future attacks from that threat actor.
  • Who?  Attack attribution is always challenging, just because you can geo-locate an attacker's IP adjust doesn't mean that's the source of the attack - they can be pivoting through other compromised hosts and networks, or they could be utilising some form of anonymising network, such as The Onion Router (TOR).  Was this an external party, an insider or someone from your supply chain?  If an internal party is involved, it is sometime necessary to increase the priority of the incident, due to the internal knowledge of assets and provisioned access to internal resources exponentially increasing the potential expediency of attack execution and likelihood of success.  Threat Intelligence can be used to profile the Tools, Techniques & Procedures (TTPs) that specific groups use and this can be compared to what you are seeing - again, the decision to continue to observe, or immediately escalate can be driven by the potential impact on the business of continuing to allow the threat actor access to your resources, compared with the benefit of gaining greater accuracy in attribution and confidence in understanding their capabilities.  
  • How?  Intrusion analysis should allow you to build a timeline of the attack.  For instance, which users (if any) were involved?  Which systems were compromised?  What data was accessed?  Again, this timeline can be compared with Threat Intelligence to identify a potential modus operandi that can be used to attribute the attack to a particular group, which will give additional insight into intent and motivation.
  • What?  What assets have been compromised?  What data has been accessed?    
  • Why?  This is ultimately the most difficult question to answer.  What was the ultimate intent of the attacker?  As I've mentioned previously, if you can utilise known TTPs to attribute the attack to a specific group, you can look at what the outcome and intent of other attacks by that group was and use this to inform why they would be attacking your organisation. Otherwise you may have to guess based on what you're seeing in the attack, such as the nature of the assets they are accessing.  The decision to continue to observe rather than immediately contain/eradicate should be taken based on the potential risk, but the use of tools such as honeypots and tarpits can allow your analysts to observe behaviour without risk to live production systems and data.

Even in those organisations that do drive their Use Cases from attempting to model an attack, they often have a tendency to gravitate towards using logs such as firewall, anti-virus and Intrusion Detection/Prevention Systems (IDS/IPS).  Why is this? Well that because these logs are typically from devices that are owned and configured by the Information Security Engineering team, rather than IT Operations.  In many organisations the relationship between IT Operations, whose main focus is on continuity and performance of services, and Information Security, whose focus on the security of data, can be challenging.  

Logs from firewall, anti-virus and IDS/IPS systems are becoming increasing less relevant in modern attacks.  Attackers today will attempt to make their initial exploitation attempts, command & control and exfiltration look as much like normal day-to-day traffic as possible - the traditional ACCEPT/DENY rules of a firewall based on IP address and port of source and destination just doesn't cut-the-mustard anymore.  Even 'next generation' firewalls cannot provide the level of visibility in encrypted packets over Secure Sockets Layer (SSL) for instance.   Just like next generation firewalls, IDS/IPS cannot look for signatures of known attacks on networks where the traffic is encrypted by technologies such as Secure Sockets Layer (SSL) without the break-out of the encryption.  Finally, traditional anti-virus solutions have been notorious for not stopping modern attacks.  It's fairly simple to look at the vendor's names listed technical skills of current and ex-employees of your organisation on LinkedIn.  Then it's just a case of finding an exploit that the vendor's product can't find using a resource like VirusTotal and you've got your in.

SOC Mistake #8: You don’t speak the language of business, you speak the language of security

This is by far one of the most common failings of Security Operations.  I’ve reviewed the maturity of several large global Security Operations Centres and they appear to be doing a reasonable job of the prediction, detection and investigation of information security incidents – but none of this is visible to the rest of the organisation who funds their operational budgets.

It is common to find someone who has started life as an operational information security person, maybe originally a firewall or Intrusion Detection System administrator, who’s career ultimately takes them to SOC Manager.  Their life has been steeped in the operational reports produced by technical controls such as firewalls, Intrusion Detection Systems and anti-virus solutions.  These reports a meaningful to him, although I’d argue about the contextual value you can gain out of an individual control’s report, but count based metrics such as ’37 unauthorised access attempts across the business’ or ‘300,000 block spam emails’ are pretty meaningless to senior management.

I’m reminded of Monty Python’s Spanish Inquisition Scene set in Jarrow in 1911 where Graham Chapman enters and says to the mill owner: “One on’t cross beams gone owt askew on treddle” to which the mill owner, both unaccustomed to the regional dialect and technical jargon says “Pardon?“.

Graham Chapman’s character is looking to the mill owner for support and direction, but he’s presenting the problem in the operational language he understands.  If he’d said “A vital piece of the manufacturing equipment in our rail sleeper production has become mis-aligned halting production” instead of “I didn’t expect some kind of Spanish Inquisition” the problem would have got sorted and Cardinals Ximinez, Fang and Biggles would have never appeared.
It is the job of the Security Operations Centre, just like the rest of the information security function, to present meaningful decision support management information around information risk to the management and it’s the management’s responsibility to make decisions on risk based on it.  If we never provide information in a language or format they can utilise, we’re always going to be seen as these strange people who live in the basement and occasionally come into the boardroom and start speaking something that sounds like Klingon to the C-level execs/

The other issue is that information security risk isn’t the only risk that companies need to consider, even if we do try to treat ourselves like a special little snowflake.  An organisation’s risk function has to balance – liquidity risk; currency risk; supply chain risk; asset risk; competition risk; pricing risk; and capital availability to name but a few.  We’re often so wrapped up in our own little worlds that are so important to want we do and we vent when decisions don’t go “our-way” forgetting that the C-level suite are running a company whose main business probably isn’t information security.
A classic example I can illustrate happened to myself.  I came into a budget meeting armed with a risk assessment and a budgeted control suite – risk to the business was around 800K, cost of controls were about 200K and residual risk would have been around 200K.  Job done the average quantitative risk wonk would say (it was a much more in-depth risk analysis than I am demonstrating here for the sake of brevity), but the issue was that if that 200K was invested in three more sales people they’d bring in much more than 800K in revenue to the organisation, which at the time was exhibiting now’t wrong with a Yorkshire accenta 98% customer retention rate, and future growth, and operational costs, was funded out of customer subscriptions.  When you took into the balance of information security risk vs. opportunity risk, my project was a bad call.

So the presentation of risk in a language the business understands and that allows a normalised comparison with other forms of risk, if you operate an Enterprise Risk Management framework, is one of the key success criteria for good security operations.

So what does good management information look like?  Well, financial metrics are a good start.

Everyone in the C-suite understand pounds, shillings and pence (excuse my pre-metric example, dollars and cents to my US friends) – it’s a good place to start.  Creating financial metrics has long been a difficult proposition, but there are several ways to do it.  Myself, I tend to map my SIEM event categories onto the VERIS framework.  This then lets me use the average costs and time-to-resolve metrics from the Verizon Data Breach Investigations Report, which I still consider to be one of the best yardsticks of what is going on in the wider world, to show my organisations performance against the average.

The other is that it most have context, providing count based metrics for the whole organisation doesn’t impart any information about what line-of-business assets are involved and what the potential bottom-line is to the business.  “37 unauthorised access attempts across Acme Corp” says one thing, “34 on cardholder processing systems”, “2 on bank transfer systems”, “1 on the customer relationship management system”, all of which are buried deep inside your infrastructure behind several layers of now’t wrong with a Yorkshire accentnow’t wrong with a Yorkshire accentcontrols, says quite another.  I’m going to talk more about this in a later blog posting, so I am going to part this here for a while.

Another aspect of the context is granularity, and this normally requires input from the analysts and incident responders and some form of established taxonomy for the more granular categorisation of incidents.  For instances saying you’ve blocked “34 malware infections” says one thing, saying “24 malware infections were stopped at the host level and we’re detected by the Intrusion Detection System”, or out of the other 10, “5 exhibited DNS behaviour showing they attempted to connect over port 443 to external systems” and that the other “5 encrypted the harddisks of systems in our payroll department just before payday”.

It’s not just about the granularity, it’s also about the curation.  Helping the exec’s understand what the impact is to the business is?; giving advice on what they could do about it?  who is the likely perpetrator is, based on the tools, techniques and procedures they are using, or at least provide an indicator of what their capability is?; understanding when this started, is it a part of a campaign or a single attack? Is it still ongoing?; how did this occur?  What vulnerabilities did the attacker exploit?  How could this be prevented from happening again?; the most difficult, and often most important question: why did this attacker attack us?  What were they after?

Having management information that allows information security risk to have a seat at the boardroom table with the rest of the functions that handle risk is a starting point. now’t wrong with a Yorkshire accent Providing context to the C-level execs enabling them to make informed risk decisions helps move security operations from a reactive function, to one that is proactive.  When this is coupled with the topics I’m going to talk about in my next couple of postings: providing line-of-business metrics and using threat intelligence, we’re moving from a Jarrow-accent to a Received Pronunciation one – although there’s now’t wrong with a Yorkshire accent as my fiancée is from Hull.

SOC Mistake #9: You don’t tier your SOC staff

Security Information and Event Management (SIEM) platforms are all about turning the mass of raw events that occur in your organisation’s infrastructure into intelligence that can be assessed by analysts and incident responders to identify and react to information security incidents.

SIEMs, despite what vendors will tell you, are not magic.  It will take you months to tune your ruleset to eliminate the bulk of false positives and you’re probably working against a moving target of an increasing number of event sources as well as continually having to adjust the rules to detect the new threats you’re facing.

To ensure the maximum use of your highly-skilled trained analysts, it is common to tier your analysts into at least two layers.

The initial layer that are solely responsible (at least to start with) for the triage of incoming events.  That is the identification of false positives, ensuring the appropriate prioritisation and escalation.

In an effective SOC, however, these level 1 analysts are not simply “click-monkeys”, as well as triaging false positives they should be doing some form of initial assessment so they can evaluate the potential impact and scope of the incident.  They should also be performing some form of adversary characterisation by evaluating where in the attack chain the event was detected (further down the chain, such as at the command or control or lateral movement stage,  may imply that they have conducted significant reconnaissance and have crafted a specific exploit to be undetectable to your host or network Intrusion Detection/Protection System – this implies a motivated and fairly skilled adversary) and they should also be, from their initial investigation, ascertaining the potential impact to the business.

Often the SIEM will have some form of prioritisation algorithm based on a number of factors, but only a human analysts can take all of the context into consideration (Skill level of attacker? Does the attacker exhibit known behaviour in their Tools, Techniques and Procedures (TTP) that can assist with attribution?  What is the apparent intent of the attacker (disruption, theft, espionage)?  Is this a one-off event or part of a sustained campaign?  Does the attack demonstrate investment of a lot of time or funds (use of zero days, for instance)? What systems are effected and what line-of-business do they support?

Only events assessed as what the level 1 analyst deals real events are escalated to the next level of more skilled analysts to conduct a deeper level of investigation.  You can create specialisations at the Level 2, or above, layers to allow workflows to be created that direct events of a certain category to specific analysts, or groups of analysts.  Some organisations have as many as three or four tiers of analysts, gradually becoming more skills and specialised as you move up the chain.

Any false positives discovered by the analysts can be routed to content authors who can further tune the SIEM rules to try and prevent the false positive from occurring in the future.
The focus should be on making this process as efficient and repeatable as possible, while allowing the collection of metrics to support continual improvement.  For instance, in HP ArcSight, we create ActiveLists for a ‘triage channel’ and the ‘content needs tuning’.  As we’re largely automating this workflow we can collect metrics on key operational Key Performance Indicators such as time-to-triage, time-to-investigate, number of false positives per use case category, number of events escalated per analyst, number of incorrectly categorised false positives per analyst.  These metrics, when combined together, can help you achieve the right balance of efficiency and effectiveness.

In my practice we've evaluated hundreds of Security Operations Centres were all of the analysts are highly trained and all operate at a single tier.  They all randomly pick the events they wish to work on off the console and do their typical ‘deep dive’ investigation.  This causes several problems:

  1. It’s hard to maintain but a broad-spectrum of investigatory skills needed for triage of all event types and a deep-level of specialisation to do a full investigation;
  2. The analyst may prefer to investigate specific categories of events, meaning that some event types may remain in the triage channel for extended periods of time;
  3. Having your highly-skilled analysts conduct the initial triage of false-positives is a bad use of their time; and
  4. Often Security Operations Centres find it really difficult to produce meaningful metrics on the overall performance of the capability, or individual analysts.

Implementing at least a two-tier system of triage/prioritisation and investigation can dramatically increase the performance of your Security Operations Centres.


SOC Mistake #10: You confuse your SOC with your NOC

Network Operations Centres (NOCs) are responsible for the operational monitoring of infrastructure and services. Their function is to identify, investigate, prioritise and escalate/resolve issues that could, or do, effect performance or availability. A Security Operation Centre (SOC) shares much in common with a NOC, it’s function is to identify, investigate, prioritise and escalate/resolve issues that could, or do, effect the security of an organisation’s information assets.