Israeli Software

Assuming you knew why a data breach will happen, wouldn’t you take your best shot at preventing it?

Consider this:

Your security defenses don’t improve your understanding of the root causes of data breaches, and without understanding the root causes –  your best shot is not good enough.

Why is this so?

First of all – defenses are by definition, not a means of improving our understanding of strategic threats. Think about the Maginot Line in WWI or the Bar-Lev line in 1973. Network and application security products that are used to defend the organization are rather poor at helping us understand and reduce the operational risk of insecure software.

Second of all – it’s hard to keep up.  Security defense products have much longer product development life cycles then the people who develop day zero exploits. The battle is also extremely asymmetric – as it costs millions to develop a good application firewall that can mitigate an attack that was developed at the cost of three man months and a few Ubuntu workstations. Security signatures (even if updated frequently) used by products such as firewalls, IPS and black-box application security are no match for fast moving, application-specific source code vulnerabilities exploited by attackers and contractors.

Remember – that’s your source code, not Microsoft.

Third – threats are evolving rapidly. Current defense in depth strategy is to deploy multiple tools at the network perimeter such as firewalls, intrusion prevention and malicious content filtering. Although content inspection technologies such as DPI and DLP are now available, current focus is primarily on the network, despite the fact that the majority of attacks are on the data – customer data and intellectual property.

The location of the data has become less specific as the notion of trusted systems inside a hard perimeter has practically disappeared with the proliferation of cloud services, Web 2.0 services, SSL VPN and convergence of almost all application transport to HTTP.

Obviously we need a better way of understanding what threats really count for our business. More about that in some up coming posts.

Now it’s some lazy journalist at Information Week aiding and abetting the pseudo-statistics of of the Ponemon Institute – screaming headlines of  the cost of data breaches of PHI – protected healthcare information

According to Information Week; Analysis: Healthcare Breach Costs May Reach $800 Million

Since the Health Information Technology for Economic and Clinical Health Act or HITECH Act of 2009 came to being, a number of new privacy, security and reporting and non-compliance penalty provisions went into effect. And as summarized by this report from HITRUST, there have been 108 entities who have reported security breaches since September of last year.

Those breaches comprise about 4 million people and records.

In the analysis, Chris Hourihan Manager, CSF Development and Operations, HITRUST used the 2009 Ponemon Institute Cost of a Data Breach Study [.pdf], which found the average cost for each record within a data breach to be $204. That’s $144 of indirect costs and $60 of direct costs. An overview of the Ponemon study is available here.

What is the connection between the Ponemon studies (sponsored by data security vendors) and the PHI leakages.

Nothing.

Why is a PII leak and a meaningless plug number of $60 relevant to PHI (which requires a combination of medical data and personal identifiers?

Why can’t someone make a phone call and ask how much the companies actually paid in fines and then make a few more phone calls and start estimating ancillary costs and direct costs such as legal.

Why not just multiply by the average cost of an iPhone?

After all you can steal data with your mobile easily enough can’t you.

Aug 7, 2010 WASHINGTON, D.D.—U.S. Senators Mark Pryor (D-AR) and John D. (Jay) Rockefeller IV (D-WV) today introduced legislation to require businesses and nonprofit organizations that store consumers’ personal information to put in place strong security features to safeguard sensitive data, alert consumers when this data has been breached, and provide affected individuals with the tools they need to protect their credit and finances. Currently, there is no single federal standard for guarding many types of consumer information.

I cannot believe my eyes – “no single federal standard”??

I am at a loss to understand why the US needs another data security bill – when there are already a plethora of regulations regarding personal information – Graham Leach Bliley (financial services), PCI DSS (credit cards), HIPAA (health care) and the state data security bills (CA SB 1386, Mass Data privacy etc.. ).  This is without even mentioning FISMA and the NIST security requirements for implementing HIPAA. With Obamacare in effect – it seems to me that the gold standard for PII protection will soon become HIPAA and since health care appears to becoming nationalized in the US – NIST will soon be the king of data security control frameworks.

Looking at data security  as an exercise in providing cost effect security countermeasures, it appears to me that the bill is most likely either a public relations play  or congressional logrolling. The interesting item is the requirement to provide credit card monitoring services after a breach for a year – perhaps the bill is intended to help stimulate the business of companies like Experian, Symantec, RSA and Mcafee.

The US does not need more data security regulation (requiring “strong security features” whatever that means) because with over 350 million US credit cards breached – the data is already out there. This bill is equivalent to closing the barn door after the horses have already fled.

What I would recommend to the esteemed Senators is a totally different approach – one adopted by Poland. Poland, which is a member of the EU and subject to the EU Privacy Law decided a few years back to make data security breaches expensive. If a firm in Poland breaches personal data – they are liable to up to a 2.5% fine of their annual gross revenue.

None of this hokey – “provide monitoring services and notify within 60 days” nonsense. Make US data breachers pay for their security vulnerabilities and even the playing field with the consumers – who are indeed paying the price for poor data security at American retailers and banks.

Are we in the same valley of death that held  content management applications in the 90s?  Where companies spent 6-7 figures on content management from companies like Vignette and over 50% of the projects never got off the ground?

Tell me what you think in this Linked In poll – DLP success or failure

It seems that with amorphous and rapidly evolving trend of storing data in cloud providers and social media like Twitter and Facebook, that social media and cloud computing is the next frontier of data security breaches.

And – here, we have not even solved the problem of trusted insiders.

The letter of the law is always operative and the common denominator of the regulators (HIPAA, PCI etc..) is not to store or transmit personal information at all in the application software systems.

We are correct in identifying cloud providers as a potential vulnerability – however, storing data in the ‘cloud’ is no different from storing data in an outsourced data center and it’s subsequent exposure to employees, outsourcing contractors etc..If you have a medical file application,  ecommerce or an online application – your best data security countermeasure is NOT to store PII at all in your application.

I personally don’t buy into technology silver bullets and data obfuscation as effective security countermeasures.   They have their utility but even if the data is obfuscated in the cloud it still traverses some interface between the data provider and the cloud provider.

In my experience, since almost all data breaches occur on the interface – adding an additional technology layer will serve to increase your value at risk not reduce it – since more complexity and more third party software only adds additional vulnerabilities and increases your threat surface.

As far as I know, there have been no documented events of PII being leaked from an infrastructure cloud provider like Rackspace or IBM. Their standards of operation and security are far better than the average business.

Notwithstanding legal definitions, regulatory standards like HIPAA and SOX tell us to do a top down risk analysis and demonstrate why the risk of leaking PII is acceptably low.

If you are developing and maintaining an online application with patient or customer data, your best bet is good application engineering and resolving your data privacy exposure issues by simply removing ePHI and PII from your systems.

If you deploy or are considering data security technology from Websense, Fidelis, Verdasys , Guardium, Imperva or Sentrigo – do you give a DAM ?

It seems that DLP (data loss prevention)  vendors are moving up the food chain into DAM (database activity monitoring)? As customers deploy two products in parallel (for example Imperva and Fidelis) for DLP and DAM – the opportunity for reducing TCO (total cost of ownership) seems to be a clear imperative.

Both Websense and Fidelis Security  provide DLP functionality for structured data in databases (Fidelis calls it internal DLP) and Websense provides fairly granular fingerprinting of combinations of relational table columns using their PreciseID technology.

Although Websense focuses on deep content analysis and stays away from application security, Verdasys provides application logging at the end point and Fidelis provides application analysis via the network session in addition to the deep content inspection. Both are functions strongly related to database activity monitoring.

Here are the goals I would put down for database activity monitoring, due to the high level of interaction with client/sever and Web applications

The  key vulnerabilities of a business  to fraud and data loss are rooted in the  four sins of hubris: thinking, looking, fighting and denying.

Hubris is defined as excessive pride or self-confidence, starting with the thought that fraud and data theft won’t happen to you.  Most firms look in the wrong direction, by focussing on external threats and malware instead of trusted insiders and organized crime. They fight the wrong battle, by installing anti-virus on machines that are not vulnerable to virus attacks, and relying on firewalls for data loss prevention. By not monitoring outbound data flows they also gain plausible denial that there are issues of data loss and economic crime in the organization.

The  sins of hubris lead to a situation where the bigger you are the harder you fall (“It can’t happen to me because we have governance, IT etc..”). According to PWC 2009 Global Economic Crime Survey – bigger companies experienced more fraud.

46% of organisations experiencing economic crime had more than 1,000 employees.

The percentage of companies in the 201 – 1,000 employee range experienced almost half the number of fraud of their larger cousins. But this may be because they have fewer governance programmes in place, or what they do have are less effective.

By the way, I think the PwC have it wrong.   Smaller companies may have fewer governance programs in place, and because they have less money, these programs are probably more effective, not less effective.

Denial of data loss and economic crime also derives from incomplete understanding of the economic costs. The 2009 PwC economic crime survey points out that :

27% of those reporting fraud in the last 12 months put its costs at more than $500,000.

One quarter of respondents reporting accounting fraud estimated that it had cost them more than US$1m.

Only 17% of those who suffered asset misappropriation reported losses of more than US$1m.

The impact of economic crime is not just financial: 32% of respondents said employee morale was most affected by such incidents.

Data loss and fraud events are unpredictable, high impact events without precedent that cannot be forecasted with virus/epidemiology or  market risk models.  The assumption in these  models is that the unexpected can be predicted by extrapolating trends from past observations, especially when these statistics are assumed to represent samples from a normal distribution. Although other distributions might provide better fits to historical data, such as the fractal (for earthquakes) or LÉvy distributions (for securities returns) or EVT (for operational risk events) – in all economic crime cases, organizational  culture was at the center of losses, and more specifically, a complex interaction of culture, people and rapidly-changing technology.

It’s impossible to stave off fraud and data theft with technology or procedures alone due the complexity, but with a management that puts a priority on a business objective of protecting company assets and customers, an organization will be able to go beyond governance and security checklists and reduce their value at risk.

Economic crime and data theft  warrants a zero-tolerance culture starting in the boardroom and with the executive management leading by example with open doors and ethical behavior.

Best practices for data security are still evolving – as there are no industry-standard data security metrics and a confusing array of regulatory compliance and industry standards – PCI DSS 1.2, Sarbanes-Oxley, FISMA, ISO2700x – just to name a few.

Organizations (government included) currently use a combination of tactics – penetration testing, vulnerability analysis (usually at the network and sometimes at the application software layer), “fire and forget” compliance exercises and technology countermeasures such as IPS/IDS, network DLP, agent DLP, database firewalls, encryption on demand, Web application firewalls.

The one countermeasure I have never seen is standardized screening.  Borrowing an approach from health-care, consider the following:

Standardized screening for suicide risk in primary care can detect adolescents with suicidal ideation, allowing referral to a behavioral healthcare center before a fatal or serious suicide attempt is made, according to the results of a study reported online April 12 and published in the May print issue of Pediatrics.

“Several associations and federal agencies have called for depression screening in pediatric primary care,” writes Matthew B. Wintersteen, PhD, from Thomas Jefferson University in Philadelphia, Pennsylvania. “Screening for suicide risk is a natural adjunct to this call….To our knowledge, this is the first study to prospectively examine the impact of standardized screening for suicide risk on detection and referral rates in pediatric primary care.”

The goals of the study were to evaluate whether brief standardized screening for suicide risk in pediatric primary care practices could improve detection of youth with suicidal ideation, maintain improved rates of detection and referral, and be duplicated in other practices.

It seems to me that duplicating brief standardized screening to data security practice is eminently possible.   A possible approach would involve using a standard threat model based on a comprehensive set of security controls – (ISO 27001 would work fine for this purpose).  The process would start with a pre-screening preparation exercise that an organization could do in the office in 1-2 hours.   After the preparation exercise, a group of 3-5 people from a business unit would meet with a data security specialist for the standardized screening that would walk through the threat model and gauge probability of occurrence of vulnerabilities and  percent damage to assets by threats.  Based on my experience, this sort of walk-through would take 2-3 hours using the structured threat model.  The result of the threat analysis would be a level of value at risk to the organization for data security and indeed a 1/2 day qualifies as brief enough.

How does your company mitigate the risk of data security threats?

Is your company management adopting a policy of “It’s other peoples money”?

In a recent thread on LinkedIn - Jody Keyser shared some quotes from David Vose’s book on risk, reliability and computerized risk modeling:  Risk Analysis a quantitative guide.

The responses to correctly identified and evaluated risks are many but generally fall into one of the following categories:

- Cancel Project
- Eliminate ( do it another way)
- Transfer (insure back to back contract)
- Share (with partner or contractor )
- Reduce (take a less risky approach)
- Add a contingency (increase budget, deadline etc.,to allow for possibility of risk)
- Collect more data to better understand risk
- Do nothing (cost is just too dang high)
- Increase ( maybe the plan is too cautious )

In my experience – when it comes to data security, data loss prevention, DLP projects – the top 2 responses to data security threats are “accept the risk” followed by “cancel the project” in a close second place.

The other alternatives are almost all non-starters. The question is – why?

Eliminating risk by changing the business process is often not an option or too much trouble for employees. For example – consider the process of transferring documents to external contractors – even though it’s trivial to encrypt documents inside a Zip file and share the password – most companies don’t make it part of their security procedure and those that do require encryption of documents sent to external business partners, don’t deploy DLP monitoring to ensure compliance with the encryption policy.

There are multiple reasons for data security risk being accepted by business managers.  Most are related to cost, complexity, changing business requirements and a tacit disbelief in effectiveness of technology in preventing data theft and fraud.

The reasons for accepting data security risk are related to  the difference between being secure and feeling secure.  Since most companies don’t monitor data flows, they don’t know how many sensitive digital assets are being leaked to the competition – ergo they don’t have the empirical data to analyze their data security threats and measure data security risks in terms of dollar threat to the business.  This would lead to enable a business to deploy data security countermeasures and be secure at an acceptable cost. It would also enable them to measure the cost effectiveness of their data security technology and challenge their innate beliefs and skepticism.

However – the company management already feel secure because they have delegated that part of  the business to the information security folks and reading the papers tells them that customers (not the business management) pay the cost of a data security breach.

As a kid growing up in South Jersey – when there was the occasional report of an urban boondoggle or million dollar NASA toilets – my Dad (who worked for RCA on defense projects and knew about these things) would always use the expression – “Other peoples money” or if it was closer to home – “Pa’s rich and Ma don’t care”…which is really close to home this year for Americans as President Obama takes the US to an unprecedented $1.35 trillion budget deficit in  2010.

One of the famous canons in the Jewish Passover “seder” ritual is 4 questions from 4 sons – the son who is wise, the son who is wicked, the son who is innocent and the son who doesn’t know enough to ask.

I sometimes have this feeling of Deja vu when considering data security technology solutions. Although the analogy is not at all parallel – I have written a list of 4 questions to be asked when considering a DLP solution – these questions require clear, authoritative answers just like in the Passover seder (להבדיל).

1. What is the key threat scenario?
2. How much Value at Risk is on the table?
3. Who owns the project?
4. Does the DLP technology fit the threat scenario?

1 – What is the key threat scenario?

Here are some typical threat scenarios – the key threat scenario should keep a C-level executive awake at night.

See http://www.software.co.il/wordpress/2010/02/is-there-a-business-need-for-dlp/

2 – What is your value at risk?

Once you have identified the key threat scenario, you must know how much value at risk is generated when a threat exploits vulnerabilities to cause damage to assets. The basis for measuring VaR (value at risk) is the asset value (generally determined by the CFO) -

VaR = asset value x threat probability x estimated damage to asset value in a percentage

The VaR is reduced by a set of security countermeasures that also have a cost. VaR is best calculated in a data security based risk assessment that uses DLP technology to measure frequencies of threat occurrence and a calculative threat model to derive VaR.

Most companies are not at a sufficient level of security maturity to do this exercise themselves – and will need an independent consultant with specific data security expertise and the ability to do analytical threat modeling.

Within a couple weeks, you should be able to get a picture of your current data security events, know your data value at risk in Euro and build a prioritized program for cost-effective DLP countermeasures.

See http://www.software.co.il/wordpress/2010/01/building-a-business-case-for-dlp/

3 – Who owns the project?

Beware of organizational politics and silos and conflicting agendas.  Need I say more?

4 – Does the DLP technology fit the threat scenario?

Just because the vendor sold you an anti-virus product doesn’t mean that his DLP technology is a good fit (even if it’s free)

Example A:  A network DLP solution may be required with 1GB throughput, if the technology saturates at 200MB/S then the solution is not a good fit.

Example B:  An agent DLP solution may be required that is capable of identifying IP in AutoCAD files; if the content analysis software is incapable of decoding AutoCAD, then the countermeasure does not mitigate the vulnerability.