Developing Security Requirements

I've just managed to get a set of "security requirements" agreed by my large financial client.

So why write requirements at all?

In a complex organisation where there are many system delivery departments with complex commercial arrangements including outsourcers and contracting organisations - writing a definitive set of standards for designing & developing applications can be

• too prescriptive,
  
• expensive - because telling 3rd party delivery organisations to "do it my way" is always expensive
• inflexible for the business.

Since mandating the implementation would be a risky endeavour - I've created requirements at a higher level that each of the organisational departments can take and use it to derive their own:

• training and awareness processes
• coding standards and quality gateways
• testing approaches and automatic review
  

Clearly, some of the departments will be more efficient by co-operating but some of the 3rd party suppliers will simply wind these requirements into their own delivery mechanisms. My clients audit staff then have a clear set of requirements to measure delivery projects and also measure the 3rd party suppliers themselves.

Rating and Prioritisation
My plan involved taking the OWASP Top Ten list, the PCI DSS standards and the FSA Treating Customer Fairly legislature and turning them into requirements; with each sub requirement rated with a MoSCoW rating. The MoSCoW rating system is standard for requirements in that it not only prioritises requirements but also allows the author to set custom measurement:

• Must Have: means that it the system does not meet this security requirement then the project has to stop as soon as it is known - the project cannot move forward.
• Should Have: means that the system has to design in to meet this requirement. If it cannot meet the requirement then the project can only go ahead if there is business, tech & risk agreement that the project can carry on.
• Could Have: means that the project have got to consider the requirement in the plans, but may drop it if it is too expensive. They have to give a reason and any risk exposure explained - but this is not scruitinised in the same way as a should have.
• Want, or Would Have: Aspirational security requirements that the business recognise as important but understand they will cost more, or might not be possible with todays technology etc.

I find it surprising that many security professionals that I deal with do not recognise RISK. They automatically will say YES, WE MUST HAVE all the security features without realising that the selection of the security features is a risk balance.

e.g. one organisation I worked for mandated fibre to every desktop because the risk or ethernet radiation leakage would be too damaging for their business. Clearly desktops in a financial insurance company do not require this - although, if the price was right then the business would have this extra security. So for one organisation it was a MUST, the other a WOULD HAVE.

What about the requirements that have been missed?
Sure, there will be some that have been missed. However, in my experience, the online applictaions that will generate huge breaches tend to show signs very early on that anti-patterns existed. Over-worked teams, poorly educated, poor communication etc etc,

By focussing the requirements on the OWASP etc, it gives auditors concrete foundations to measure the security quality of projects in a commercial context.

Example Requirement
ID: Z1.1 (in the AntiXSS chapter)
Name: Strong output encoding
Description: Ensure that all user-supplied data is appropriately entity encoded before rendering, taking the approach to encode all characters other than a very limited subset. This is the approach of the various Anti-XSS libraries. Also, set the character encodings for each page output, which will reduce exposure to some attack vectors.
Type: Security requirement
Source: OWASP Top 10, 2007
MoSCoW: MUST
Measure of Success: All input and output data must be encoded appropriately with the output encoding specified, e.g. UTF-1, ISO 8859-1