A good model should be able to explain, but a good scientific model should be able to predict and a good model for our purposes should help us design as well. At a minimum, the ability to predict leads to the ability to design by random variation and selective survival with the survival evaluation being made based on prediction. In most cases, it is a lot more efficient to have the ability to create design rules that are reflective of some underlying structure.
Any model we build that is to have utility must be computationally reasonable relative to the task at hand. Far more computation is likely to be available for a large-scale strategic deception than for a momentary tactical deception, so it would be nice to have a model that scales well in this sense. Computational power is increasing with time, but not at such a rate that we will ever be able to completely ignore computational complexity in problems such as this.
A fundamental design problem in deception lies in the fact that deceptions are generally thought of in terms of presenting a desired story to the target, while the available techniques are based on what has been found to work. In other words, there is a mismatch between available deception techniques and technologies and objectives.
Rather than focus on what we wish to do, our approach is to focus on what we can do and build up 'deception programs' from there. In essence, our framework starts with a programming language for human deception by finding a set of existing primitives and creating a syntax and semantics for applying these primitives to targets. We can then associate metrics with the elements of the programming language and analyze or create deceptions that optimize against those metrics.
The framework for human deception then has three parts:
A set of primitive techniques: The set of primitive techniques is extensive and is described hierarchically based on the model shown above, with each technique associated with one or more of Observables, Actions, Assessments, Capabilities, Expectations, and Intent and causing an effect on the situation depicted by the model.
Properties of those techniques: Properties of techniques are multi-dimensional and include all of the properties discussed in this report. This includes, but is not limited to, resources consumed, effect on focus of attention, concealment, simulation, memory requirements and impacts, novelty to target, certainty of effect, extent of effect, timeliness of effect, duration of effect, security requirements, target system resource limits, deceiver system resource limits, the effects of small changes, organizational structure, knowledge, and constraints, target knowledge requirements, dependency on predisposition, extent of change in target mind set, feedback potential and availability, legality, unintended consequences, the limits of modeling, counterdeception, recursive properties, and the story to be told. These are the same properties of deception discussed under "The Nature Of Deception" earlier.
A syntax and semantics for applying and optimizing the properties: This is a language that has not yet been developed for describing, designing, and analyzing deceptions. It is hoped that this language and the underlying database and simulation mechanism will be developed in subsequent efforts.
The astute reader will recognize this as the basis for a computer language, but it has some differences from most other languages, most fundamentally in that it is probabilistic in nature. While most programming languages guarantee that when you combine two operators together in a sequence you get the effect of the first followed by the effect of the second, in the language of deception, a sequence of operators produces a set of probabilistic changes in perceptions of all parties across the multi-dimensional space of the properties of deception. It will likely be effective to "program" in terms of desired changes in deception properties and allow the computer to "compile" those desired changes into possible sequences of operators. The programming begins with a 'firing table' of some sort that looks something like the following table, but with many more columns filled in and many more details under each of the rows. Partial entries are provided for technique 1 which, for this example, we will choose as 'audit suppression' by packet flooding of audit mechanisms using a distributed set of previously targeted intermediaries.
Deception Property | Technique 1 | ... | Technique n |
---|---|---|---|
name | Audit Suppression | ||
general concept | packet flooding of audit mechanisms | ||
means | using a distributed set of intermediaries | ||
target type | computer | ||
resources consumed | reveals intermediaries which will be disabled with time | ||
effect on focus of attention | induces focus on this attack | ||
concealment | conceals other actions from target audit and analysis | ||
simulation | n/a | ||
memory requirements and impacts | overruns target memory capacity | ||
novelty to target | none - they have seen similar things before | ||
certainty of effect | 80% effective if intel is right | ||
extent of effect | reduces audits by 90% if effective | ||
timeliness of effect | takes 30 seconds to start | ||
duration of effect | until ended or intermediaries are disabled | ||
security requirements | must conceal launch points and intermediaries | ||
target system resource limits | memory capacity, disk storage, CPU time | ||
deceiver system resource limits | number of intermediaries for attack, pre-positioned assets lost with attack | ||
the effects of small changes | nonlinear effect on target with break point at effectiveness threshold | ||
organizational structure and constraints | Going after known main audit server which will impact whole organization audits | ||
target knowledge | OS type and release | ||
dependency on predisposition | Must be proper OS type and release to work | ||
extent of change in target mind set | Large change - it will interrupt them - they will know they are being attacked | ||
feedback potential and availability | Feedback apparent in response behavior observed against intermediaries and in other fora | ||
legality | Illegal except at high intensity conflict - possible act of war | ||
unintended consequences | Impacts other network elements, may interrupt other information operations, may result in increased target security | ||
the limits of modeling | Unable to model overall network effects | ||
counterdeception | If feedback known or attack anticipated, easy to deceive attacker | ||
recursive properties | only through counter deception | ||
possible deception story | We are concealing something - they know this - but they don't know what |
Considering that the total number of techniques is likely to be on the order of several hundred and the vast majority of these techniques have not not been experimentally studied, the level of effort required to build such a table and make it useful will be considerable.
For a moment, we will pause from the general issue of deception and examine more closely the situation of an attacker attempting to exploit a defender through information system attack. In this case, there is a commonly used attack methodology that subsumes other common methodologies and there are only three known successful attack strategies identified by simulation and verified against empirical data. We start with some background.
The pathogenesis of diseases has been used to model the process of breaking onto computers and it offers an interesting perspective. [63] In this view, the characteristics of an attack are given in terms of the survival of the attack method.
|
|
|
This particular perspective on attack as a biological process ignores one important facet of the problem, and that is the preparation process for an intentional and directed attack. In the case of most computer viruses, targeting is not an issue. In the case of an intelligent attacker, there is generally a set of capabilities and an intent behind the attack. Furthermore, survival (stability in the environment) would lead us to the conclusion that a successful attacker who does not wish to be traced back to their origin will use an intelligence process including personal risk reduction as part of their overall approach to attack. This in turn leads to an intelligence process that precedes the actual attack.
The typical attack methodology consists of:
There are loops from higher numbers to lower numbers so that, for example, privilege expansion can lead back to intelligence and system entry or forward to subversion, and so forth. In addition, attackers have expectations throughout this process that adapt based on what has been seen before this attack and within this attack. Clean up, observation of effects, and analysis of feedback for improvement are also used throughout the attack process.
Extensive simulation has been done to understand the characteristics of successful attacks and defenses. [5] Among the major results of this study were a set of successful strategies for attacking computer systems. It is particularly interesting that these strategies are similar to classic military strategies because the simulation methods used were not designed from a strategic viewpoint, but were based solely on the mechanisms in use and the times, detection, reaction, and other characteristics associated with the mechanisms themselves. Thus the strategic information that fell out of this study was not biased by its design but rather emerged as a result of the metrics associated with different techniques. The successful attack strategies identified by this study included:
Slow, loud attacks tend to be detected and reacted to fairly easily. A successful attacker can use combinations of these in different parts of an attack. For example, speed can be used for a network scan, stealth for system entry, speed for privilege expansion and planting of capabilities, stealth for verifying capabilities over time, and overwhelming force for exploitation. This is a typical pattern today.
Substantial red teaming and security audit experience has led to some speculations that follow the general notions of previous work on individual deception. It seems clear from experience that people who use computers in attacks:
If this turns out to be true, it has substantial implications for both attack and defense. Experiments should be undertaken to examine these assertions as well as to study the combined deception properties of small groups of people working with computers in attacking other systems. Unfortunately, current data is not adequate to thoroughly understand these issues. There may be other strategies developed by attackers, other attack processes undertaken, and other tendencies that have more influence on the process. We will not know this until extensive experimentation is done in this area.
From the deceptive defender's perspective, there also seem to be a limited set of strategies.
Computer Only: If the computer is being used for a fully automated attack, analysis of the attack tool or relatively simply automated response mechanisms are highly effective at maintaining the computer's expectations, dazzling the computer to induce unanticipated processing and results, feeding false information to the computer, or in some cases, causing the computer to crash. We have been able to easily induce or suppress signal returns to an attacking computer and have them seen as completely credible almost no matter how ridiculous they are. Whether this will continue and to what extent it will continue in the presence of a sophisticated hostile environment remain to be seen.
People Only: Manual attack is very inefficient so it is rarely used except in cases where very specific targets are involved. Because humans do tend to see what they expect to see, it is relatively easy to create high fidelity deceptions by redirecting traffic to a honey pot or other such system. Indeed, this transition can even be made fairly early in an attack without most human attackers noticing it. In this case there are three things we might want to do:
(1) maintain the attackers expectations to consume their time and effort,
(2) slowly change their expectations to our advantage at a rate that is not noticeable by typical humans (e.g., slow the computer's response minute by minute till it is very slow and the attacker is wasting lots of time and resources), and
(3) create cognitive dissonance to force them to think more deeply about what is going on, wonder if they have been detected, and induce confusion in the attacker.
People With Poorly Integrated Computers: This is the dominant form of efficient widespread attack today. In this form, people use automated tools combined with short bursts of human activity to carry out attacks.
The intelligence process is almost entirely done by scanning tools which (1) can be easily deceived and (2) tend to be believed. Such deceptions will only be disbelieved if inconsistencies arise between tools, in which case the tools will initially be suspected.
System entry is either automated with the intelligence capability or automated at a later time when the attacker notices that an intelligence sweep has indicated a potential vulnerability. Results of these tools will be believed unless they are incongruous with normal expectations.
Privilege expansion is either be fully automated or has a slight manual component to it. It typically involves the loading of a toolkit for the job followed by compilation and/or execution. This typically involves minimal manual effort. Results of this effort are believed unless they are incongruous with normal expectations.
Planting capabilities is typically nearly automated or fully automated. Returning to verify over time is typically automated with time frames substantially larger than attack times. This will typically involve minimal manual effort. Results of this effort will be believed unless they are incongruous with normal expectations.
Exploitation is typically done under one-shot or active control. A single packet may trigger a typical exploit, or in some cases the exploit is automatic and ongoing over an extended period of time. This depends on whether speed, stealth, or force is desired in the exploitation phase. This causes observables that can be validated by the attacker. If the observables are not present it might generate deeper investigation by the attacker. If there are plausible explanations that can be discovered by the attacker they will likely be believed.
People With Well Integrated Computers: This has not been observed to date. People are not typically augmenting their intelligence but rather automating tasks with their computers.
As in the case with attacker strategies, few experiments have been undertaken to understand these issues in detail, but preliminary experiments appear to confirm these notions.
Several authors have written simplistic analyses and provided rules of thumb for deception planning. There are also some notions about planning deceptions under the present model using the notions of low, middle, and high level cognition to differentiate actions and create our own rules of thumb with regard to our cognitive model. But while notions are fine for contemplation, scientific understanding in this area requires an experimental basis.
According to [10] a 5-step process is used for military deception. (1) Situation analysis determines the current and projected enemy and friendly situation, develops target analysis, and anticipates a desired situation. (2) Deception objectives are formed by desired enemy action or non-action as it relates to the desired situation and friendly force objectives. (3) Desired [target] perceptions are developed as a means to generating enemy action or inaction based on what the enemy now perceives and would have to perceive in order to act or fail to act - as desired. (4) The information to be conveyed to or kept from the enemy is planned as a story or sequence, including the development and analysis of options. (5) A deception plan is created to convey the deception story to the enemy.
These steps are carried out by a combination of commander and command staff as an embedded part of military planning. Because of the nature of military operations, capabilities that are currently available and which have been used in training exercises and actual combat are selected for deceptions. This drives the need to create deception capabilities that are flexible enough to support the commander's needs for effective use of deceptions in a combat situation. From a standpoint of information technology deceptions, this would imply that, for example, a deceptive feint or movement of forces behind smoke screens with sonic simulations of movement should be supported by simulated information operations that would normally support such action and concealed information operations that would support the action being covered by the feint.
Deception maxims are provided to enhance planner understanding of the tools available and what is likely to work: [10]
Magruder's principles - the exploitation of perceptions: It is easier to maintain an existing belief than to change it or create a new one.
Limitations of human information processing: The law of small numbers (once you see something twice it is taken as a maxim), and susceptibility to conditioning (the cumulative effect of small changes). These are also identified and described in greater detail in Gilovich [14].
Cry-Wolf: This is a variant on susceptibility to conditioning in that, after a seeming threat appears again and again to be innocuous, it tends to be ignored and can be used to cover real threats.
Jones' Dilemma: Deception is harder when there are more information channels available to the target. On the other hand, the greater the number of 'controlled channels', the better it is for the deception.
A choice among deception types: In "A-type" deception, ambiguity is introduced to reduce the certainty of decisions or increase the number of available options. In "M-type" deception, misdirection is introduced to increase the victim's certainty that what they are looking for is their desired (deceptive) item.
Axelrod's contribution - the husbanding of assets: Some deceptions are too important to reveal through their use, but there is a tendency to over protect them and thus lose them by lack of application. Some deception assets become useless once revealed through use or overuse. In cases where strategic goals are greater than tactical needs, select deceptions should be held in reserve until they can be used with greatest effect.
A sequencing rule: Sequence deceptions so that the deception story is portrayed as real for as long as possible. The most clear indicators of deception should be held till the last possible moment. Similarly, riskier elements of a deception (in terms of the potential for harm if the deception is discovered) should be done later rather than earlier so that they may be called off if the deception is found to be a failure.
The importance of feedback: A scheme to ensure accurate feedback increases the chance of success in deception.
The Monkey's Paw: Deceptions may create subtle and undesirable side effects. Planners should be sensitive to such possibilities and, where prudent, take steps to minimize these effects.
Care in the designed and planned placement of deceptive material: Great care should be used in deceptions that leak notional information to targets. Apparent windfalls are subjected to close scrutiny and often disbelieved. Genuine leaks often occur under circumstances thought improbable.
Deception failures are typically associated with (1) detection by the target and (2) inadequate design or implementation. Many examples of this are given. [10]
As a doctrinal matter, Battlefield deception involves the integration of intelligence support, integration and synchronization, and operations security. [10]
Intelligence Support: Battlefield deceptions rely heavily on timely and accurate intelligence about the enemy. To make certain that deceptions are effective, we need to know (1) how the target's decision and intelligence cycles work, (2) what type of deceptive information they are likely to accept, (3) what source they rely on to get their intelligence, (4) what they need to confirm their information, and (5) what latitude they have in changing their operations. This requires both advanced information for planning and real-time information during operations.
Integration and Synchronization: Once we know the deception plan we need to synchronize it with the true combat operations for effect. History has shown that for the greatest chance of success, we need to have plans that are: (1) flexible, (2) doctrinally consistent with normal operations, (3) credible as to the current situation, and (4) simple enough to not get confused during the heat of battle. Battlefield deceptions almost always involve the commitment of real forces, assets, and personnel.
Operations Security: OPSEC is the defensive side of intelligence. In order for a deception to be effective, we must be able to deny access to the deceptive nature of the effort while also denying access to our real intentions. Real intentions must be concealed, manipulated, distorted, and falsified though OPSEC.
In the DoD context, it must be assumed that any enemy is well versed in DoD doctrine. This means that anything too far from normal operations will be suspected of being a deception even if it is not. This points to the need to vary normal operations, keep deceptions within the bounds of normal operations, and exploit enemy misconceptions about doctrine. Successful deceptions are planned from the perspective of the targets.
The DoD has defined a set of factors in deceptions that should be seriously considered in planning [10]. It is noteworthy that these rules are clearly applicable to situations with limited time frames and specific objectives and, as such, may not apply to situations in information protection where long-term protection or protection against nebulous threats are desired.
Policy: Deception is never an end in itself. It must support a mission.
Objective: A specific, realistic, clearly defined objective is an absolute necessity. All deception actions must contribute to the accomplishment of that objective.
Planning: Deception should be addressed in the commander's initial guidance to staff and the staff should be engaged in integrated deception and operations planning.
Coordination: The deception plan must be in close coordination with the operations plan.
Timing: Sufficient time must be allowed to: (1) complete the deception plan in an orderly manner, (2) effect necessary coordination, (3) promulgate tasks to involved units, (4) present the deception to the enemy decision-maker through their intelligence system, (5) permit the enemy decision maker to react in the desired manner, including the time required to pursue the desired course of action.
Security: Stringent security is mandatory. OPSEC is vital but must not prevent planning, coordination, and timing from working properly.
Realism: It must look realistic.
Flexibility: The ability to react rapidly to changes in the situation and to modify deceptive action is mandatory.
Intelligence: Deception must be based on the best estimates of enemy intelligence collection and decision-making processes and likely intentions and reactions.
Enemy Capabilities: The enemy commander must be able to execute the desired action.
Friendly Force Capabilities: Capabilities of friendly forces in the deception must match enemy estimates of capabilities and the deception must be carried out without unacceptable degradation in friendly capabilities.
Forces and Personnel: Real forces and personnel required to implement the deception plan must be provided. Notional forces must be realistically portrayed.
Means: Deception must be portrayed through all feasible and available means.
Supervision: Planning and execution must be continuously supervised by the deception leader. Actions must be coordinated with the objective and implemented at the proper time.
Liaison: Constant liaison must be maintained with other affected elements to assure that maximum effect is attained.
Feedback: A reliable method of feedback must exist to gage enemy reaction.
Deception of humans and automated systems involves interactions with their sensory capabilities. [10] For people, this includes (1) visual (e.g., dummies and decoys, camouflage, smoke, people and things, and false vs. real sightings), (2) Olfactory (e.g., projection of odors associated with machines and people in their normal activities at that scale including toilet smells, cooking smells, oil and gas smells, and so forth), (3) sonic (e.g., directed against sounding gear and the human ear blended with real sounds from logical places and coordinated to meet the things being simulated at the right places and times) (4) electronic (i.e., manipulative electronic deception, simulative electronic deception, and imitative electronic deception).
Resources (e.g., time, devices, personnel, equipment, materiel) are always a consideration in deceptions as are the need to hide the real and portray the false. Specific techniques include (1) feints, (2) demonstrations, (3) ruses, (4) displays, (5) simulations, (6) disguises, and (7) portrayals. [10]
A typical deception is carried out by the creation and invocation of a deception plan. Such a plan is normally based on some set of reasonably attainable goals and time frames, some understanding of target characteristics, and some set of resources which are made available for use. It is the deception planner's objective to attain the goals with the provided resources within the proper time frames. In defending information systems through deception our objective is to deceive human attackers and defeat the purposes of the tools these humans develop to aid them in their attacks. For this reason, a framework for human deception is vital to such an undertaking.
All deception planning starts with the objective. It may work its way back toward the creation of conditions that will achieve that objective or use that objective to 'prune' the search space of possible deception methods. While it is tempting for designers to come up with new deception technologies and turn them into capabilities; (1) Without a clear understanding of the class of deceptions of interest, it will not be clear what capabilities would be desirable; and (2) Without a clear understanding of the objectives of the specific deception, it will not be clear how those capabilities should be used. If human deception is the objective, we can begin the planning process with a model of human cognition and its susceptibility to deception.
The skilled deception planner will start by considering the current and desired states of mind of the deception target in an attempt to create a scenario that will either change or retain the target's state of mind by using capabilities at hand. State of mind is generally only available when (1) we can read secret communications, (2) we have insider access, or (3) we are able to derive state of mind from observable outward behavior. Understanding the limits of controllable and uncontrollable target observables and the limits of intelligence required to assure that the target is getting and properly acting (or not acting) on the information provided to them is a very hard problem.
Deception Levels
In the model depicted above and characterized by the diagram below, three levels can be differentiated for clearer understanding and grouping of available techniques. They are characterized here by mechanism, predictability, and analyzability:
Level | Mechanism | Predictability | Analysis | Summary |
---|---|---|---|---|
Low-level | Low-level deceptions operate at the lower portions of the areas labeled observables and actions. They are designed to cause the target of the deception to be physically unable to observe signals or to cause the target to selectively observe signals. | Low-level deceptions are highly predictable based on human physiology and known reflexes. | Low-level deceptions can be analyzed and very clearly characterized through experiments that yield numerical results in terms of parameters such as detection thresholds, response times, recovery times, edge detection thresholds, and so forth. | Except in cases where the target has sustained physiological damage, these deceptions operate very reliably and predictably. The time frames for these deceptions tend to be in the range of milliseconds to seconds and they can be repeated reliably for ongoing effect. |
Mid-Level | Mid-level deceptions operate in the upper part of the areas labeled Observables and Actions and in the lower part of the areas marked Assessment and Capabilities. They are generally designed to either: (1) cause the target to invoke trained or pattern matching based responses and avoid deep thought that might induce unfavorable (to us) actions; or (2) induce the target to use high level cognitive functions, thus avoiding faster pattern matching responses. | Mid-level deceptions are usually predictable but are affected by a number of factors that are rather complex, including but not limited to socialization processes and characteristics of the society in which the person was brought up and lives. | Analysis is based on a substantial body of literature. Experiments required for acquiring this knowledge are complex and of limited reliability. There are a relatively small number of highly predictable behaviors. These relatively small number of behaviors are common and are invoked under predictable circumstances. | Many mid-level deceptions can be induced with reasonable certainty through known mechanisms and will produce predictable results if applied with proper cautions, skills, and feedback. Some require social background information on the subject for high surety of results. The time frame for these deceptions tends to be seconds to hours with lasting residual effects that can last for days to weeks. |
High-level | High-level deceptions operate from the upper half of the areas labeled Assessment and Capabilities to the top of the chart. They are designed to cause the subject to make a series of reasoned decisions by creating sequences of circumstances that move the individual to a desired mental state. | High-level deceptions are reasonably controlled if adequate feedback is provided, but they are far less certain to work than lower level deceptions. The creation and alteration of expectations has been studied in detail and it is clearly a high skills activity where greater skill tends to prevail. | High-level deception requires a high level of feedback when used against a skilled adversary and less feedback under mismatch conditions. There is a substantial body of supporting literature in this area but it is not adequate to lead to purely analytical methods for judging deceptions. | High level deception is a high skills game. A skilled and properly equipped team has a reasonable chance of carrying out such deceptions if adequate resources are applied and adequate feedback is available. These sorts of deceptions tend to operate over a time frame of hours to years and in some cases have unlimited residual effect. |
Deception Guidelines
This structuring leads to general guidelines for effective human deception. In essence, they indicate the situations in which different levels of deception should be used and rules of thumb for their use.
Low-Level |
- Higher certainty can be achieved at lower levels of perception.
|
Mid-Level |
- If a low-level deception will not work, a mid-level
deception must be used.
|
High-Level |
- If the target cannot be forced to make a mid-level
decision in your favor, a high-level deception must be used.
|
Just as Sun Tzu created guidelines for deception, there are many modern pieces of advice that probably work pretty well in many situations. And like Sun Tzu, these are based on experience in the form of anecdotal data. As someone once said: The plural of anecdote is statistics.
As more and more of these sorts of rules of thumb based on experience are combined with empirical data from experiments, it is within the realm of plausibility to create more explicit algorithms for decision planning and evaluation. Here is an example of the codification of one such algorithm. It deals with the issue of sequencing of deceptions with different associated risks identified above.
Let's assume you have two deceptions, A (low risk) and B (high risk). Then, if the situation is such that the success of either means the mission is accomplished, the success of both simply raises the quality of the success (e.g. it costs less), and the discovery of either by the target will increase the risk that the other will also fail, then you should do A first to assure success. If A succeeds you then do B to improve the already successful result. If A fails, you either do something else or do B out of desperation. On the other hand, if the situation is such that the success of both A and B are required to accomplish the mission and if the discovery of either by the target early in execution will result in substantially less harm than discovery later in execution, then you should do B first so that losses are reduced if, as is more likely, B is detected. If B succeeds, you then do A. Here this is codified into a form more amenable to computer analysis and automation:
GIVEN: Deception A (low risk) and Deception B (high risk). IF [A Succeeds] OR [B Succeeds] IMPLIES [Mission Accomplished, Good Quality/Sched/Cost] AND [A Succeeds] AND [B Succeeds] IMPLIES [Mission Accomplished, Best Quality/Sched/Cost] AND [A Discovered] OR [B Discovered ] IMPLIES [A (higher risk) AND B (higher risk)] THEN DO B [comment: Do high-risk B first to insure minimal loss in case of detection] IF [B Succeeds] DO A (Late) [comment: Do low-risk A second to improve outcome] ELSE DO Out #1 [comment: Do higher-risk A because you're desperate.] OR ELSE DO Out #n [comment: Do something else instead.] IF [A Succeeds] OR [B Succeeds] IMPLIES [Mission Accomplished, Good Quality/Sched/Cost] AND [A Detected] OR [B Detected] IMPLIES [Mission Fails] AND [A Discovered Early] OR [B Discovered Early] IMPLIES [Mission Fails somewhat] AND [A Discovered Late] OR [B Discovered Late] IMPLIES [Mission Fails severely] THEN DO B [comment: Do high-risk B first to test and advance situation] IF [B Early Succeeds] DO A (Late) [comment: Do low-risk A second for max chance of success] IF [A Late Succeeds (likely)] THEN MISSION SUCCEEDS. ELSE [A Late Fails (unlikely)] THEN MISSION FAILS/in real trouble. ELSE [B Early Fails] [Early Failure] DO Out #1 [comment: Do successful retreat as pre-planned.] OR DO Out #m [comment: Do another pre-planned contingency instead.]
We clearly have a long way to go in codifying all of the aspects of deception and deception sequencing in such a form, but just as clearly, there is a path to the development of rules and rule-based analysis and generation methods for building deceptions that have effect and reduce or minimize risk, or perhaps optimize against a wide range of parameters in many situations. The next reasonable step down this line would be the creation of a set of analytical rules that could be codified and experimental support for establishing the metrics associated with these rules. A game theoretical approach might be one of the ways to go about analyzing these types of systems.