Research

I’m very fortunate to have inherited Rothman hair, which is gray but plentiful and grows fast. Like fungus. Given my schedule, I tend to wait until things get lost in my hair before I get it cut. Like birds; or yard debris; or Nintendo DS games. A few weeks back the Boss told me to get it cut when I lost my iPhone in my hair. So I arranged a day to hit the barber I have frequented for years. I usually go on Mondays when I can, because his partner is off. These guys have a pretty sophisticated queuing system, honed over 40+ years. Basically you wait until your guy is open. That works fine unless the partner is open and your guy is backed up. Then the partner gives me the evil eye as he listens to his country music. But I have to stay with my guy because he has a vacuum hooked up to his clipper. Yes, I wait for my guy because he uses a professional Flowbee. But when I pulled up the shop was closed. I’ve been going there for 7 years and the shop has never been closed on Monday. Then I looked at the sign, which shows hours only for the partner – my guy’s hours aren’t listed. Rut roh, I got a bad feeling. But I was busy, so I figured I’d go back later in the week and see what happened. I went in Thursday, and my guy wasn’t there. Better yet, the partner was backed up, but I had just lost one of the kids in my hair, so I really needed a cut. I’m quick on the uptake, so I figured something was funky, but all my guy’s stuff was still there – including pictures of his grandkids. It’s like the place that time forgot. But you can’t escape time. It catches everyone. Finally the situation was clarified when a customer came in to pay his respects to the partner. My fears were confirmed: my guy was gone, his trusty clippers silenced. The Google found his obituary. Logically I know death completes the circle of life, and no one can escape. Not even my barber. Truth be told, I was kind of sad. But I probably shouldn’t be. Barber-man lived a good life. He cut hair for decades and enjoyed it. He did real estate as well. He got a new truck every few years, so the shop must have provided OK. He’d talk about his farm, which kept him busy. I can’t say I knew him well, but I’m going to miss him. So out of respect I wait and then sit in the partner’s chair. Interestingly enough he gave me a great cut, even though I was covered in hair without the Flowbee. I was thinking I’d have to find a new guy, but maybe I’ll stick with partner-man. Guess there is a new barber-man in town. Godspeed Richard. Enjoy the next leg of your journey. -Mike Photo credits: “Barber Shop” originally uploaded by David Smith Incite 4 U Can I call you Dr. Hacker?: Very interesting analysis here by Ed Moyle about whether security should be visionary. Personally I don’t know what that means, because our job is to make sure visionary business leaders can do visionary things without having critical IP or private data show up on BitTorrent. But the end of the post on whether security will be innovation-driven (like product development), standards-driven, innovation-averse (like accounting), or standard-driven, innovation-accepting (like medicine) got me thinking. I think we’d like to think we’ll be innovation-driven, but ultimately I suspect we’ll end up like medicine. Everyone still gets sick (because the viruses adapt to our defenses), costs continue to skyrocket, and the government eventually steps in to make everything better. Kill me now, Dr. Hacker. – MR Learn clarity from the (PHP)Fog: One of the things that fascinates me about breaches (and most crisis events) is how the affected react. As I wrote about last week, most people do almost exactly the wrong thing. But as we face two major breaches within our industry, at RSA (“everyone pretend you don’t know what’s going on even though it’s glaringly obvious”), and Comodo (“we were the victim of a state-sponsored attack from Iran, not a teenager, we swear”); perhaps we should learn some lessons from PHPFog (“How We Got Owned by a Few Teenagers (and Why It Will Never Happen Again)”). Honesty is, by far, the best way to maintain the trust of your customers and the public. Especially when you use phrases like, “This was really naive and irresponsible of me.” Treat your customers and the public like adults, not my 2 year old. Especially when maintaining secrecy doesn’t increase their security. – RM MySQL PwNaGe: For the past few days, the news that mysql.com has both a SQL injection vulnerability and a Cross Site Scripting (XSS) vulnerability has been making the rounds. The vulnerabilities are not in the MySQL database engine, but in the site itself. Detailed information from the hacked site was posted on Full Disclosure last Sunday as proof. Appearently the MySQL team was alerted to the issue in January, and this looks like a case of “timely disclosure” – they could have taken the hack further if they wanted. Not much in takeaways from this other than SQL injection is still a leading attack vector and you should have quality passwords to help survive dictionary attacks in the aftermath of a breach. Still no word from Oracle, as there is no acknowledgement of the attack on mysql.com. I wonder if they will deploy a database firewall? – AL APT: The FUD goes on and on and on and on: I applaud Chris Eng’s plea for the industry to stop pushing the APT FUD at all times. He nails the fact that vendors continue to offer solutions to the APT because they don’t want to miss out when the “stop APT project” gets funded. The nebulous definition of APT helps vendors obfuscate the truth, and as Chris points out it frustrates many of us. Yes, we should call out vendors for

In our introduction to Security Benchmarking, Going Beyond Metrics, we spent some time defining metrics and pointing out that they have multiple consumers, which means we need to package and present the data to these different constituencies. As you’ll see, there is no lack of things to count. But in reality, just because you can count something doesn’t mean you should. So let’s dig a bit into what you can count. Disclaimers: we can only go so deep in a blog series. If you are intent on building a metrics program, you must read Andy Jaquith’s seminal work Security Metrics: Replacing Fear, Uncertainty and Doubt. The book goes into great detail about how to build a security metrics program. The first significant takeaway is how to define a good security metric in the first place: Expressed as numbers Have one or more units of measure Measured in a consistent and objective way Can be gathered cheaply Have contextual relevance Contextual relevance tends to be the hard thing. As Andy says in his March 2010 security metrics article in Information Security magazine: “the metrics must help someone–usually the boss–make a decision about an important security or business issue.” That’s where most security folks tend to fall down, focusing on things that don’t matter, or drawing suspect conclusions from operational data. For example, generating a security posture rating from AV coverage won’t work well. Consensus Metrics We also need to tip our hats to the folks at the Center for Internet Security, who have published a good set of starter security metrics, built via their consensus approach. Also take a look at their QuickStart guide, which does a good job of identifying the process to implement a metrics program. Yes, consensus involves lowest common denominators, and their metrics are no different. But keep things in context: the CIS document provides a place to start, not the definitive list of what you should count. Taking a look at the CIS consensus metrics: Incident Management: Cost of incidents, Mean cost of incidents, Mean incident recovery cost, Mean time to incident discovery, Number of incidents, Mean time between security incidents, Mean time to incident recovery Vulnerability Management: Vulnerability scanning coverage, % systems with no severe vulnerabilities, Mean time to mitigate vulnerabilities, Number of known vulnerabilities, Mean cost to mitigate vulnerabilities Patch Management: Patch policy compliance, Patch management coverage, Mean time to patch, Mean cost to patch Configuration Management: % of configuration compliance, Configuration management coverage, current anti-malware compliance Change Management: Mean time to complete changes, % of changes with security review, % of changes with security exceptions Application security: # of applications, % of critical applications, Application risk access coverage, Application security testing coverage Financial: IT security spending as % of IT budget, IT security budget allocation Obviously there are many other types of information you can collect – particularly from your identity, firewall/IPS, and endpoint management consoles. Depending on your environment these other metrics may be important for operations. We just want to provide a rough sense of the kinds of metrics you can start with. For those gluttons for punishment who really want to dig in we have built Securosis Quant models that document extremely granular process maps and the associated metrics for Patch Management, Network Security Operations (monitoring/managing firewalls and IDS/IPS), and Database Security. We won’t claim all these metrics are perfect. They aren’t even supposed to be – nor are they all relevant to all organizations. But they are a place to start. And most folks don’t know where to start, so this is a good thing. Qualitative ‘Metrics’ I’m very respectful of Andy’s work and his (correct) position regarding the need for any metrics to be numbers and have units of measure. That said, there are some things that aren’t metrics (strictly speaking) but which can still be useful to track, and for benchmarking yourself against other companies. We’ll call these “qualitative metrics,” even though that’s really an oxymoron. Keep in mind that the actual numbers you get for these qualitative assessments isn’t terribly meaningful, but the trend lines are. We’ll discuss how to leverage these ‘metrics’/benchmarks later. But some context on your organization’s awareness and attitudes around security is critical. Awareness: % of employees signing acceptable use policies, % of employees taking security training, % of trained employees passing a security test, % of incidents due to employee error Attitude: % of employees who know there is a security group, % of employees who believe they understand threats to private data, % of employees who believe security hinders their job activities We know what you are thinking. What a load of bunk. And for gauging effectiveness you aren’t wrong. But any security program is about more than just the technical controls – a lot more. So qualitatively understanding the perception, knowledge, and awareness of security among employees is important. Not as important as incident metrics, so we suggest focusing on the technical controls first. But you ignore personnel and attitudes at your own risk. More than a few security folks have been shot down because they failed to pay attention to how they were perceived internally. Again, entire books have been written about security metrics. Our goal is to provide some ideas (and references) for you to understand what you can count, but ultimately what you do count depends on your security program and business imperatives. Next we will focus on how to collect these metrics systematically. Because without your own data, you can’t compare anything. Share:

Define Needs The first step in the process is to determine your needs, keeping in mind that there are two main drivers for File Activity Monitoring projects, and it’s important to understand the differences and priorities between them: Entitlement management Activity monitoring Most use cases for FAM fall into one of these two categories, such as data owner identification. It’s easy to say “our goal is to audit all user access to files”, but we recommend you get more specific. Why are you monitoring? Is your primary need security or compliance? Are there specific business unit requirements? These answers all help you pick the best solution for your individual requirements. We recommend the following process for this step: Create a selection committee: File Activity Monitoring initiatives tend to involve three major technical stakeholders, plus compliance/legal. On the IT side we typically see security and server and/or storage management involved. This varies considerably, based on the size of the organization and the complexity of the storage infrastructure. For example, it might be the document management system administrators, SharePoint administrators, NAS/storage management, and server administration. The key is to involve the major administrative leads for your storage repositories. You may also need to involve network operations if you plan to use network monitoring. Define the systems and platforms to protect: FAM projects are typically driven by a clear audit or security goal tied to particular storage repositories. In this stage, detail the scope of what will be protected and the technical specifics of the platforms involved. You’ll use this list to determine technical requirements and prioritize features and platform support later. Remember that needs grow over time, so break the list into a group of high-priority systems with immediate requirements, and a second group summarizing all major platforms you may need to protect later. Determine protection and compliance requirements: For some repositories you might want strict preventative security controls, while for others you may just need comprehensive activity monitoring or entitlement management to satisfy a compliance requirement. In this step, map your protection and compliance needs to the platforms and repositories from the previous step. This will help you determine everything from technical requirements to process workflow. Outline process workflow and reporting requirements: File Activity Monitoring workflow varies by use. You will want to define different workflows for entitlement management and activity monitoring, as they may involve different people; that way you can define what you need instead of having the tool determine your process. In most cases, audit, legal, or compliance, have at least some sort of reporting role. Different FAM tools have different strengths and weaknesses in their management interfaces, reporting, and internal workflow, so think through the process before defining technical requirements to prevent headaches down the road. By the end of this phase you should have defined key stakeholders, convened a selection team, prioritized the systems to protect, determined protection requirements, and roughed out process workflow. Formalize Requirements This phase can be performed by a smaller team working under the mandate of the selection committee. Here, the generic needs from phase 1 are translated into specific technical features, and any additional requirements are considered. This is the time to come up with criteria for directory integration, repository platform support, data storage, hierarchical deployments, change management integration, and so on. You can always refine these requirements after you begin the selection process and get a better feel for how the products work. At the conclusion of this stage you will have a formal RFI (Request For Information) for vendors, and a rough RFP (Request For Proposals) to clean up and formally issue in the evaluation phase. Evaluate Products As with any products, it can be difficult to cut through the marketing materials and figure out whether a product really meets your needs. The following steps should minimize your risk and help you feel confident in your final decision: Issue the RFI: Larger organizations should issue an RFI though established channels and contact the leading FAM vendors directly. If you’re a smaller organization start by sending your RFI to a trusted VAR and email the FAM vendors which appear appropriate for your organization. Perform a paper evaluation: Before bringing anyone in, match any materials from the vendor or other sources to your RFI and draft RFP. Currently few vendors are in the FAM market, so your choices will be limited, but you should be fully prepared before you go into any sales situations. Also use outside research sources and product comparisons. Bring in vendors for on-site presentations and demonstrations: Instead of a generic demonstration, ask each vendor to walk through your specific use cases. Don’t expect a full response to your draft RFP – these meetings are to help you better understand the different options and eventually finalize your requirements. Finalize your RFP and issue it to your short list of vendors: At this point you should completely understand your specific requirements and issue a formal, final RFP. Assess RFP responses and begin product testing: Review the RFP results and drop anyone who doesn’t meet any of your minimal requirements (such as platform support), as opposed to ‘nice-to-have’ features. Then bring in any remaining products for in-house testing. You will want to replicate your highest volume system and its traffic if at all possible. Build a few basic policies that match your use cases, and then violate them, so you get a feel for policy creation and workflow. Select, negotiate, and buy: Finish testing, take the results to the full selection committee, and begin negotiating with your top choice. Internal Testing In-house testing is the last chance to find problems in your selection process. Make sure you test the products as thoroughly as possible. And keep in mind that smaller organizations may not have the resources or even the opportunity to test before purchase. A few key aspects to test are: Platform support and installation: Determine agent or integration compatibility (if needed) with your repositories. If you plan to use agents or integrate with a document management system, this is one

Once again, I have knocked off a series of posts for a new white paper. The title is “Understanding and Selecting a File Activity Monitoring Solution”. Although there are only a few vendors in the market, this is a technology I have been waiting a few years for, and I think it’s pretty useful. There are basically two sides to it: Entitlement management: Collecting all user privileges in monitored file repositories, linking into your directory servers, and giving you a highly simplified process for cleaning up all the messes and managing things better when moving forward. Activity monitoring: Full activity monitoring for all your file repositories in scope… including alerting for policy violations. It’s pretty cool stuff – imagine setting a policy to alert you any time someone copies an entire directory off the server instead of a single file. Or copying 30 files in a day, when they normally only open 1 or 2. And that’s just scratching the surface of the potential. The links to all the posts are below, and I could use any feedback you have before we convert this puppy to a paper and post it. (If you are seeing this in RSS, you will have to click the post to see all the links, because I’m too lazy to add them in manually). Share:

The Ponemon Institute has released a white paper, What auditors think about Crypto (registration required). I downloaded and took a cursory look at their results. My summary of their report is “IT auditors rely on encryption, but key management can be really hard”. No shock there. A client passed along a TechTarget blog post where Larry Ponemon is quoted as saying auditors prefer encryption , but worded to make their study sound like a comparison between encryption and tokenization. So I dove deep into their contents to see if I missed something. Nope. The study does not compare encryption to tokenization, and Larry’s juxtaposition implies it is. The quotes from the TechTarget post are as follows: Encryption has always been a coveted technology to auditors, but organizations that have problems with key management may view tokenization as a good alternative and Tokenization is an up and coming technology; we think PCI DSS and some other compliance requirements will allow tokenization as a solid alternative to encryption. and In general auditors in our study still favor encryption in all the different use cases that we examined, Which are all technically true but misleading. If you had to choose one technology over another for all use cases, I don’t know of a security professional who wouldn’t choose encryption, but that’s not a head to head comparison. Tokenization is a data replacement technology; encryption is a data obfuscation technology. They serve different purposes. Think about it this way: There is no practical way for tokenization to protect your network traffic, and it would be a horrible strategy for protecting backup tapes. You can’t build a VPN with tokenization – the best you could do would be to use access tokens from a Kerberos-like service. That does not mean tokenization won’t be the best way to secure data at rest security now or in the future. Acknowledging that encryption is essential sometimes and that auditors rely on it is a long way from establishing that encryption is better or preferable technology in the abstract. Larry’s conclusion is specious. Let’s be clear: the vast majority of discussion around tokenization today has to do with credit card replacement for PCI compliance. The other forms of tokens used for access and authorization have been around for many years and are niche technologies. It’s just not meaningful to compare cryptography in general against tokenization within PCI deployments. A meaningful comparison of popularity between encryption and tokenization, would need to be confined to areas where they can solve equivalent business problems. That’s not GLBA, SOX, FISMA, or even HIPAA; currently it’s specifically PCI-DSS. Note that only 24% of those surveyed were PCI assessors. They look at credit card security on a daily basis, and compare relative merits of the two technologies for the same use case. 64% had over ten years experience, but PCI audits have been common for less than 5. The survey population is clearly general auditors, which doesn’t seem to be an appropriate audience for ascertaining the popularity of tokenization – especially if they were thinking of authorization tokens when answering the survey. Of customers I have spoken to, who want to know about tokenization, more than 70% intend to use tokenization to help reduce the scope of PCI compliance. Certainly my sample size is smaller than the Ponemon survey’s. And the folks I speak with are in retail and finance, so subject to PCI-DSS compliance. At Securosis we predict that tokenization will replace encryption in many PCI-DSS regulated systems. The bulk of encryption installations, having nothing to do with PCI-DSS and being inappropriate use cases for tokenization, however, will be unchanged. At a macro level these technologies go hand in hand. But as tokenization grows in popularity, in suitable situations it will often be chosen over encryption. Note that encryption systems require some form of key management. Thales, the sponsor of Ponemon’s study, is a key vendor in the HSM space which dominates key management for encryption deployments. Finally, there is some useful information in the report. It’s worth a few minutes to review, to look get some insight into decision makers and where funding is coming from. But it’s just not possible to make a valid comparison between tokenization and encryption from this data. Share:

Now that we have covered the base features it’s time to consider how these tie in with policies, workflow, and reporting. We’ll focus on the features needed to support these processes rather than defining the processes themselves. Policy Creation File Activity Monitoring products support two major categories of policies: Entitlement (Permissions/Access Control) policies. These define which users can access which repositories and types of data. They define rules for things like orphaned user accounts, separation of duties, role/group conflicts, and other situations that don’t require real-time file activity. Activity-based polices. These alert and block based on real-time user activity. When evaluating products, look for a few key features to help with policy creation and management: Policy templates that serve as examples and baselines for building your own policies. A clean user interface that allows you to understand business context. For example, it should allow you to group categories, pool users and groups to speed up policy application (e.g., combine all the different accounting related groups into “Accounting”), and group and label repositories. This is especially important given the volume of entries to manage when you integrate with large user directories and multi-terabyte repositories. New policy wizards to speed up policy creation. Hierarchical management for multiple FAMs in the same organization. Role-based administration, including roles for super administrators and assigning policies to sub-administrators. Policy backup and restore. Workflow As with policy creation, we see workflow requirements focusing on the two major functions of FAM: entitlement management and activity monitoring. Entitlement Management This workflow should support a closed-loop process for collection of privileges, analysis, and application of policy-based changes. Your tool should do more than merely collect access rights – it should help you build a process to ensure that access controls match your policies. This is typically a combination of different workflows for different goals – including identification of orphan accounts with access to sensitive data, excessive privileges, conflict of interest/separation of duties based on user groups, and restricting access to sensitive repositories. Each product and policy will be different, but they typically share a common pattern: Collect existing entitlements. Analyze based on policies. Apply corrective actions (either building an alerting/blocking policy or changing privileges). Generate a report of identified and remediated issues. The workflow should also link into data owner identification because this must often be understood before changing rights. Activity Monitoring and Protection The activity monitoring workflow is very different than entitlement management. Here the focus is on handling alerts and incidents in real time. The key interface is the incident handling queue that’s common to most security tools. The queue lists incidents and supports various sorting and filtering options. The workflow tends to follow the following structure: Incident occurs and alert appears in the queue. It is displayed with the user, policy violated, and repository or file involved. The incident handler can investigate further by filtering for other activity involving that user, that repository, or that policy over a given time period (or various combinations). The handler can assign or escalate the incident to someone else, close the incident, or take corrective actions such as adjusting the file permissions manually. The key to keeping this efficient is not requiring the incident handler to jump around the user interface in a manual process. For example, clicking on an incident should show its details and then links to see other related incidents by user, policy, and repository. Incidents should also be grouped logically – an attempt to copy an entire directory should appear as one incident, not one incident for each of 1,000 files in the repository. Any FAM product may also include additional workflows, such as one for identifying file owners. Reporting One of the most important functions for any File Activity Monitoring product is robust reporting – this is particularly important for meeting compliance requirements. Aside from a repository of pre-defined reports for common requirements such as PCI and HIPAA, the tool should allow you to generate arbitrary reports. (We hate to list that as a requirement, but we still occasionally see security tools that don’t support creation of arbitrary reports). Share:

Introduction Our entire profession is called “information security”, but surprisingly few of our technologies focus on actually protecting the data itself, as opposed to the infrastructure surrounding it. Data Loss Prevention emerged nearly 10 years ago to address exactly this problem. By peering inside files, network traffic, and other sources – and understanding both content and context – DLP provides new capabilities comparable to when we first started looking inside network packets. The Data Loss Prevention market is split into two broad categories of tools – full suites dedicated to DLP, and what we call “DLP Light”. There is lots of confusion about the differences between these approaches… and even their definitions. In this series we will focus on DLP Light – what it is, how it works, and how to rapidly take advantage of it. (For more information on full-suite Data Loss Prevention, see our white paper Understanding and Selecting a Data Loss Prevention Solution. Defining DLP Light We are talking about a subset of Data Loss Prevention, so we need to start with our definition of DLP: Products that, based on central policies, identify, monitor, and protect data at rest, in motion, and in use, through deep content analysis. A full DLP suite includes network, storage, and endpoint capabilities; as well as a range of deep content analysis techniques such as document fingerprinting. DLP Light tools include a subset of those capabilities; they are generally features of, or integrated with, other security products – such as endpoint protection platforms, email security gateways, and next-generation firewalls. DLP Light tools tend to have some or all the following characteristics: Focused on a subset of ‘channels’. A DLP Light tool might focus on portable storage, email, web traffic, other channels, or a combination. Fewer/simpler content analysis techniques. Rather than providing a wide range of deep content analysis techniques, many of which are resource-intensive, DLP Light products tend to include a smaller set of techniques, or even a single method. The most common is pattern matching, which is the most commonly used technique in both full and Light DLP deployments. Less dedicated workflow. DLP Light tools are often integrated with, or features of, other security tools. As such, they lack the full self-contained workflow found in full DLP suites. You might ask, “So how is this still DLP?” The key defining characteristic of both full DLP and DLP Light is content analysis. If a tool can peer into network traffic or a file and sniff out something like a credit card number, it’s DLP. If all it does is rely on tagging/labeling, metadata, or contextual information… it isn’t DLP. The Role of DLP Light DLP Light plays an important role in a few different use cases: Organizations that already use the product the DLP Light tool integrates with – such as email security gateways – often want to start protecting sensitive data while constraining costs. Organizations that don’t require dedicated DLP tools. This is often due to less stringent or more circumscribed data security requirements. Organizations that want to scope out their DLP problem before investing in a dedicated tool. DLP Light can play a valuable role in helping assess data security risk. Organizations that want to start small and grow into full DLP. There is a bit of overlap between these cases, but they reflect the most common reasons we see people using DLP Light. Dedicated Data Loss Prevention is extremely powerful, but not appropriate for everyone. Next we will cover the technology side of DLP Light, and then we will finish with the Quick Wins process for rapidly deriving value from your implementation. Share:

At Securosis we tend to be passionate about security. We have the luxury of time (and lack of wingnuts yelling at us all day) to think about how security should work, and make suggestions for how to get there. We also have our own pet projects – areas of research that get us excited. We usually focus on ‘hot’ topics, because they pay the bills. We rarely get to step back and think outside the box about a security process that really needs to change. That’s why I’m very excited to be starting a new research project called Security Benchmarking, Going Beyond Metrics – interestingly enough, on security metrics and benchmarking. This topic is near and dear to my heart. I have been writing about metrics for years, and I broached the subject of benchmarking in my security methodology book (The Pragmatic CSO) back in 2007. To be candid, talking about security metrics – and more specifically security benchmarking – was way ahead of the market. Four years later, we still struggle to decide what we should count. Forget about comparing our numbers to other organizations to understand relative performance – which is how we would define a benchmark. It has been like trying to teach a toddler quantum physics. But we believe this idea’s time has come. In this series and the resulting white paper, I will revisit many of the ideas in The Pragmatic CSO, including updates based on industry progress since 2007. Ultimately, at Securosis we focus on practical (even pragmatic) application of research, so there won’t be any fluff or pie-in-the-sky handwaving. Just things you can start thinking about right now, with some actionable information to both rejuvenate your security metrics program and start comparing yourself against your peers. Before we jump in, thanks to our friends at nCircle for sponsoring this research. The rest of this series will appear on the complete (‘heavy’) side of our site and our heavy RSS feed. Introduction: Security Metrics As long as we have been doing security, we have been trying to count different aspects of our work. The industry has had vert limited success so far (yes – we are being very kind), so we need a better way to answer the question: “How effective are you at security?” The fundamental problem is that security is a nebulous topic, and at the end of the day the only important question is whether you are compromised or not – that is the ultimate measure of your effectiveness. But that doesn’t help communicate value to senior management or increase operational efficiency. The problem is further complicated by the literally infinite number of things to count. You can count emails and track which ones are bad – that’s one metric. So is the number of network flows, compared to how many of them are ‘bad’. If you can count it, it’s a metric. It may not be a good metric, but it is a metric. You can spend as much time as you like modeling, and counting, and correlating, and trying to figure out your “coverage” percentage, comparing the controls (always finite) to every conceivable attack (always infinite). But ultimately we have found that most security professionals do best keeping two sets of books. No, not like Worldcom did in the good old days, but two distinct sets of metrics: Important to senior management: Folks like the CIO, CFO, and CEO want to know whether you are ‘secure’ and how effective the security team is. They want to hear about the number of ‘incidents’, how much money you spend, and whether you hit the service levels you committed to. They tend to focus on those for ‘overhead’ functions – and whether you like it or not, security is overhead. Important to running your business: Distinct from business-centric numbers, you also need to measure the efficiency of your security processes. These are the numbers that make senior management eyes glaze over. Things like AV updates, time to re-image a machine or deploy a patch, number of firewall rule changes, and a host of other metrics that track what your folks are doing every day. The point of these numbers isn’t to gauge security quality overall, but to figure out how you can do your work faster and better. Of course, it’s almost impossible to improve things you don’t control. So we will focus on activities that can be directly impacted by the CSO and/or the security team. As we work through this series we will look at logical groupings of metrics that can be used for both operational and benchmarking purposes. But before we get ahead of ourselves, let’s define security benchmarking at a high level. Security Benchmarking Given our general failure to define and collect a set of objective, defendable measures of security effectiveness, impact, etc., a technique that can yield very interesting insight into your security environment is to compare your numbers to others. If you can get a fairly broad set of consistent data (both quantitative and qualitative), then compare your numbers to the dataset, you can get a feel for relative performance. This is what we mean by security benchmarking. Benchmarks have been used in other IT disciplines for decades. Whether for data center performance or network utilization, companies have always felt compelled to compare themselves to others. This hasn’t happened in security to date, mostly because we haven’t been sure what to count. If we can build some consensus on that, and figure out a way to collect and share that data safely, then benchmarking becomes much more feasible. Let’s discuss some metrics and why they would be interesting to compare to others: Number of incidents: Are you overly targeted? Or less effective at stopping attacks? The number of incidents doesn’t tell the entire story, but knowing how you fare relative to other is certainly interesting. Downtime for security issues: How effective you are at stopping attacks? And how severe is their impact? The downtime metric doesn’t capture everything, but it does get at the most visible impact of an attack. Number of activities: By tracking activity at a high level, you

Driven by the continued noise about the RSA and Comodo breaches, we have spent a lot of time stating the obvious this week. But then I remember that what is obvious to us may not be to everyone else. And even if it is obvious to you, sometimes you need a reminder because you are probably too busy fighting fires and answering questions from senior management (like “Don’t I take dumps in a Comodo?”) to remember the obvious stuff. So, once again, it is time to don the Captain Obvious suit and talk about layered security models. Rich reminded everyone about Crisis Communications yesterday and earlier this week; Adrian ranted about people fail trumping process fail regarding development every day of the week. Now it’s my turn. If you only have one line of defense, such as strong authentication (either two-factor or even a digital certificate) – you are doing it wrong. Yes, we still need layers in our security models. You cannot assume that any specific control will be effective, so you need a variety of controls to ensure critical information is adequately protected. That’s the underlying concept of the vaults idea balloon I have been working on. Depending on the sensitivity of the information, you layer additional controls until it’s sufficiently difficult to compromise that information. Note that I said sufficiently hard – Captain Obvious reminds us that everything can be broken. When you are building your threat models, you don’t assume your user is trusted, even after they authenticate, right? Remember, a device can be compromised after authenticating just as easily as before. Or someone could be holding a gun to your user’s head, which makes most folks pretty well willing to provide access to anything the attacker wants. That’s another reason the RSA and Comodo breaches should be business as usual. Factor in that you now have a bit less trust in the authentication layer. How does that change what you do? This is why we also advocate monitoring everything, looking for not normal, and being able to react faster and better to any situation. Yes, your controls will fail. Even when you layer them. So constantly checking for out-of-the-ordinary behavior may give you early warning that something has been pwned. In a post earlier this week, Rob Graham linked to a South Park clip where Captain Hindsight points out that BP’s critical error was not having a backup valve for the backup valve for the regular valve. Of course. And you know that in your shop Captain Hindsight will make an appearance when you get compromised. That’s part of the job, but you can make sure you are doing all you can to reduce the likelihood that one control failure will provide open access. That means don’t outside without your layers on. Photo Credit: “Captain Obvious” originally uploaded by Gareth Jones Share:

I realize that I have a tendency to overplay my emergency services background, but it does provide me with some perspective not common among infosec professionals. One example is crisis communications. While I haven’t gone through all the Public Information Officer (PIO) training, basic crisis communications is part of several incident management classes I have completed. I have also been involved in enough major meatspace and IT-related incidents to understand how the process goes. In light of everything from HBGary, to TEPCO, to RSA, to Comodo, it’s worth taking a moment to outline how these things work. And I don’t mean how they should go, but how they really play out. Mostly this is because those making the decisions at the executive level a) have absolutely no background in crisis communications, b) think they know better than their own internal experts, and c) for some strange reason tend to think they’re different and special and not bound by history or human nature. You know – typical CEOs. These people don’t understand that the goal of crisis communications is to control the conversation through honesty and openness, while minimizing damage first to the public, then second to your organization. Reversing those priorities almost always results in far worse impact to your organization – eventually, of course, the public eventually figures out you put them second and will make you pay for it later. Here’s how incidents play out: Something bad happens. The folks in charge first ask, “who knows” to figure out whether they can keep it secret. They realize it’s going to leak, or already has, so they try to contain the information as much as possible. Maybe they do want to protect the public or their customers, but they still think they should keep at least some of it secret. They issue some sort of vague notification that includes phrases like, “we take the privacy/safety/security of our customers very seriously”, and “to keep our customers safe we will not be releasing further details until…”, and so on. Depending on the nature of the incident, by this point either things are under control and there is more information would not increase risk to the public, or the attack was extremely sophisticated. The press beats the crap out of them for not releasing complete information. Competitors beat the crap out of them because they can, even though they are often in worse shape and really just lucky it didn’t happen to them. Customers wait and see. They want to know more to make a risk decision and are too busy dealing with day to day stuff to worry about anything except the most serious of incidents. They start asking questions. Pundits create more FUD so they can get on TV or in the press. They don’t know more than anyone else, but they focus on worst-case scenarios so it’s easier to get headlines. The next day (or within a few hours, depending on the severity) customers start asking their account reps questions. The folks in charge realize they are getting the crap beaten out of them. They issue the second round of information, which is nearly as vague as the first, in the absurd belief that it will shut people up. This is usually when the problem gets worse. Now everyone beats the crap out of the company. They’ve lost control of the news cycle, and are rapidly losing trust thanks to being so tight-lipped. The company trickles out a drivel of essentially worthless information under the mistaken belief that they are protecting themselves or their customers, forgetting that there are smart people out there. This is usually where they use the phrase (in the security world) “we don’t want to publish a roadmap for hackers/insider threats” or (in the rest of the world), “we don’t want to create a panic”. Independent folks start investigating on their own and releasing information that may or may not be accurate, but everyone gloms onto it because there is no longer any trust in the “official” source. The folks in charge triple down and decide not to say anything else, and to quietly remediate. This never works – all their customers tell their friends and news sources what’s going on. Next year’s conference presentations or news summaries all dissect how badly the company screwed up. The problem is that too much of ‘communications’ becomes a forlorn attempt to control information. If you don’t share enough information you lose control, because the rest of the world a) needs to know what’s going on and b) will fill in the gaps as best they can. And the “trusted” independent sources are press and pundits who thrive on hyperbole and worst-case scenarios. Here’s what you should really do: Go public as early as possible with the most accurate information possible. On rare occasion there are pieces that should be kept private, but treat this like packing for a long trip – make a list, cut it in half, then cut it in half again, and that’s what you might hold onto. Don’t assume your customers, the public, or potential attackers are idiots who can’t figure things out. We all know what’s going on with RSA – they don’t gain anything by staying quiet. The rare exception is when things are so spectacularly fucked that even the collective creativity of the public can’t imagine how bad things are… then you might want them to speculate on a worst case scenario that actually isn’t. Control the cycle be being the trusted authority. Don’t deny, and be honest when you are holding details back. Don’t dribble out information and hope it will end there – the more you can release earlier, the better, since you then cut speculation off at the knees. Update constantly. Even if you are repeating yourself. Again, don’t leave a blank canvas for others to fill in. Understand that everything leaks. Again, better for you to provide the information than an anonymous insider. Always always put your customers and the public first. If not, they’ll know