Understanding honeypots on two axes
With the release of Modern Honey Network by our friends at ThreatStream, lots of folks have started to pay attention to honeypots as a data source again. Traditionally, we have classified honeypots based on their level of interaction. Low-interaction honeypots provide a relatively small surface to an attacker, who can't do much beyond the initial contact. High-interaction honeypots, in contrast, simulate as much of a system as possible. This approach allows the attacker to take many different types of actions. Generally, the level of interaction available depends on the underlying platform (kippo, netcat, dionaea, etc.)
When deploying a honeypot or analyzing the data from it, we should also think about the focus. That is, how targeted is the scope of our data collection? A collection of sensors scattered across a bunch of different hosting providers around the world will provide a broad view with little direct relevancy to a particular enterprise. Conversely, an internally-deployed honeypot with a simulated Active Directory environment and seeded with decoy data will focus on a narrow scope of attackers within our environment.
The choices we make on each spectrum depend almost entirely on our goals for the project and the resources we can allocate. Broadly-targeted, high-interaction honeypots require an entirely different approach than narrow-targeted (internal) low-interaction honeypots. They also provide wildly different sorts of data.
The scope of a honeypot deployment depends greatly on the goals of the project. Available resources also matter greatly. Generally speaking, broadly-targeted honeypots provide unfocused data for general situational awareness. However, those with a more narrow focus allow an organization to increase their capabilities for intelligence-driven response.
Deployments should consider the risk as well. In the case of honeypots, we need to think about the possibility that an attacker could break out of the intended environment into the underlying system. This applies to any network-accessible service, but in this case the risk increases proportionally to the (intentionally) increased threat. Additional controls like system hardening and monitoring can mitigate this to an extent. The specifics largely depend on the actual honeypot platform in use.
When planning a deployment, consider the scope and focus of the intended data collection. Interaction level and targeting should serve the overall goals of the operation, whether for general awareness & research or for intelligence-driven response.