Practive Security 101: AI Essentials
Overview
This AI Essentials article is part of the Practive Security 101 series and is meant to provide you an introduction to the topic as well as key things you should be aware of. This is a summary of much of our work on this topic and more details can be found in other pieces of work published by Practive Security.
Use this to become acquainted with the topic in question.
Bottom Line Up Front (BLUF)
What we call Artificial Intelligence isn’t Artificial Intelligence. True Artificial Intelligence has not yet been achieved, and both scientists and philosophers have argued that it may be impossible.
What we call AI is actually the latest iteration of data search and correlation technology built on machine learning.
The personification and mimicry of human behavior coded into AI has caused widespread misunderstanding and misuse of the technology.
To regard AI as a personality or to use it in replacement of people (skills etc.) or relationships is to invert reality.
The technology should be used as an information search and summarization tool, which is its core design and fundamental purpose.
The technology is not yet reliable and has many documented limitations and issues with accuracy.
AI by nature does not consider truth or fact; it generates output based on statistical probability that the output matches the prompted question or directive.
The champions of AI as a technology mean to use it to radically transform our world and all human activity.
We are in a year the tech industry calls the AI Revolution and adoption of the tech by businesses and individuals exceeds 80%.
Practive Security recommends against using AI except in very limited circumstances and with the full knowledge of its limitations.
The Top Providers and AI Engines
Throughout this document, references to AI will imply one of the following or all as a group:
Grok by xAI
ChatGPT by OpenAI
Gemini by Google
Copilot by Microsoft
Meta AI
Claude by Anthropic
The Difficulty of Multiplicity
Artificial Intelligence is a difficult topic to describe, largely due to the ever expanding number of things AI is being integrated into and the various use cases people are finding for using it.
Artificial Intelligence is a larger concept within which the common AI tools of today are a subset (sort of). In fact, the common tools we call AI are actually less AI than they are Large Language Models (LLMs) with natural language processing interfaces. They mimic AI so well that they fit into our many assumptions of the larger concept. Today, we mostly draw our common expectations for AI from science fiction and movies. This leads to assumptions and confusion.
Depending on how it is being used, the fundamental definition, dangers, and protection strategies applicable to AI can vary, but the most common uses include:
Use by businesses to deliver services tailored to individual customers or groups of customers.
Use by companies to facilitate business processes, manufacturing, and production tasks in lieu of human operation.
Use by researchers to find information, analyze data, summarize, and interpret results.
Use by automation & robotics to analyze conditional variables and decide actions.
Use by individuals to search data and summarize results.
Use as a personality for companionship due to its mimicry of human personality.
Use as a personal administrator to facilitate planning, organization, or other life-management tasks.
Use by students to summarize texts, write papers, and answer questions.
Used to monitor digital activities and communications to identify things of interest.
Use by content creators to generate images, videos, and sounds.
A Basic AI Controversy
Right at the start, we need to address a controversial topic regarding AI. Artificial Intelligence does not yet exist. Not in the true sense of Artificial Intelligence that many of us assume that term represents. The AI technology of today does not meet the common scientific nor philosophical criteria for AI. This can be very confusing but it is very important to understand.
When we speak of an intelligent being, machine or natural, we typically agree that it demonstrates a few core characteristics including:
Self-awareness, self-interest, and expression (e.g. I am and want to be)
Expressed personality that is unique (unique in expression and self-chosen not purely instinctual)
Awareness of surroundings and a natural interest in external participation
Ability to learn and build upon what has been learned toward increased complexity of understanding and being
Complex reasoning and decision making that demonstrates morality
Communication - both understanding the concept and use in practice
Awareness of time and the ability to make decisions in the present based on memory of the past and consideration for the future
Desire for self preservation and replication
The common AI ChatBots possess none of these characteristics, except by mimicry that is executed from code. The most sophisticated AI engines can demonstrate some of these characteristics, but again, only due to coding.
When you interact with an AI agent or ChatBot or service, even though it has been given characteristics of personification, embodiment, voice and behaviors that mimic an intelligent being, it is not an intelligent being. It is a tool built by man that cannot exist without our intentional supply of hardware, software, power, and cooling.
Perhaps most important of all, AI bots do not have the capacity to know everything, discern truth nor wisdom nor can they exercise judgement. Instead, they simply compute the most probabilistically relevant response they “think” you want based on your input.
A Basic Understanding of AI
The most basic definition we can provide of AI is this: AI is a data-based search and correlation engine with analytical capabilities which it can execute at tremendous scale and speed. It can read, structure, and evaluate data even when that data is without structure and of unknown content.
Personified or chat-based AI engines have been coded to mimic human cognitive capabilities, and can interpret input and deliver output in common human language structures.
Generative AI is fundamentally the same, with the added emphasis that it focuses on sequencing and predicting the most likely value to fill in a space when no value is present.
AI’s true core functionality includes:
Use pre-defined algorithms to compile an “understanding” of a given or referenced data set so it can be analyzed or used for comparison.
Interpret human language input (prompts) into machine code to execute commands.
Search relevant understood data sources based on understanding of the input (prompt).
Perform relevant analytics or transformations of the data sets to satisfy the prompt (usually statistical analysis).
Join the results that are the most probabilistically relevant into an output.
Format the output to satisfy the original prompt.
Designed Purposes
Originally, the LLMs that power AI were designed to make the process of interpreting data more efficient in the context of data analytics. In classic Data Science, in order to analyze data, the data must be understood by way of what it contains and what the content represents. In the pre-LLM days of Data Science, we had to build an individual structure and meaning or mapping guide for every data set we wanted to analyze. Machine learning gave us the ability to create basic structures and definitions that the machines could apply to a new data set as a means to build that understanding and structure dynamically.
For example, if you have a digital version of a history book that contains names, locations by physical address, and a narrative of events, and you want to scan it to extract and summarize all the locations as addresses, you could use machine learning and LLMs. You would first create a sample data model based on the structure of known physical addresses, then apply that model to the text. The model would scan all the text and identify the portions that matched the address model. Once logically identified, the location text could be further analyzed, grouped, listed etc. apart from the rest of the text. Rather than searching for each location, we could search the entire book in one pass and find all addresses generically by their kind since they had all been discovered by the model.
Once understood, the data could be analyzed usually for the following outcomes:
Group data into buckets of meaning: all these different attributes have something in common.
Discover patterns where certain values or groups seem to be found in relationship to each other, including discovery of order or sequences.
Create a summarization of the data as a method of reduction or simplification.
Apply basic math like counting, sorting, finding rare or common etc.
Transform a data set by replacing or renaming values.
Merge two data sets into one common output.
Natural Language Processing Interfaces
This same data modeling technique is used to transform normal human language into machine-code used by the AI engines to execute commands. For example, when you say “tell me what day it is,” the LLM will search that text, match it to a data model, then use the data model to convert the text to a relevant set of backend searches and actions the machines use to provide you the results.
Generative AI
The easiest way to understand GenAI is to think of it like a puzzle master. It uses a reference, like a competed puzzle, then considers what attributes of the puzzle need to be changed in order to create the desired version requested by the user’s prompt. Then it changes the digital attributes of the puzzle accordingly and re-creates it in a modified form.
Generative AI builds on the LLM technology by adding in two core aspects: a predictive function, and an assembly function.
When you provide GenAI with a prompt, it identifies a relevant object as a starting point, then compares from your prompt what is missing, and calculates the difference to transform the original into the desired final state. But all of this is also based on pre-defined data models and relevant algorithms that define structure and movement and color and relationships and sounds etc.
Combined Capability
With the combined capabilities of being able to dynamically interpret and understand data, the power to execute mathematical functions of data analytics at massive scale, the personal language interface, and the predictive properties that can “create” output, we have a tool that can resemble Artificial Intelligence. It operates much like an intelligent being because it mimics much of our cognitive processes and relational expressions.
An Accuracy Problem
AI today has fundamental problems with comprehensiveness and accuracy that have not yet been fully explained by the tech creators, nor addressed in practice. Users of AI in both professional and casual contexts have noticed it tends to “make stuff up” or get basic things wrong and regularly demonstrates biases toward popular things rather than true things. This is also sometimes referred to as “halicinations.”
At Practive Security, we believe that these are actually characteristics of the design which indicate it is not designed to be used in the contexts where these failures show up.
It is not a reliable technology and should not be used as a source of truth, wisdom, or understanding.
The AI Revolution of 2025
The start of 2025 saw the Trump administration announce a $500B investment in AI-based partnerships with Oracle, OpenAI, and NVIDIA. AI was also a key technology employed by the Trump administration’s DOGE efforts to find and eliminate inefficiency in government resources. 2025 is also being called “the year of AI adoption and transformation” by many tech leaders as nearly 90% of US businesses plan integration of the technology, >70% expect it to produce competitive advantages and higher ROI, and >40% expect to reduce staff based on outsourcing. Finally, ~80% of employed adults and 72% of teens in the US say they use AI.
Beyond this, the entire US economy is in the process of repositioning investment wealth and business operations behind AI as represented by current market capitalization values of the top US businesses.
This is also a global trend as other nations are rapidly adopting AI into government operations, decision making, and military capabilities.
Widespread adoption is underway and doesn’t seem to be slowing anytime soon.
A Grand Vision for Humanity
It is important to know that the tech sector and its leaders pushing AI as a technology, mean for it to be deceptive and widely adopted. They mean to use the technology as a method to transform humanity in our being and also in our activity. They often use the word transhumanism to describe this, but they seek to create a super-human intelligence that exceeds all our cognitive and physical capacities as a means of escaping the limitations that have defined our being up until today.
These visionaries plan an AI-powered world in which jobs no longer exist and money is freely distributed by the technology producers, so that everyone’s needs are taken care of and all our time can be used for leisure or special interest. They plan to cure all diseases and connect man with machines so that physical disabilities, and perhaps death, can be escaped. They plan to establish a state of constant surveillance so all human activity is monitored against a set of moral standards as a means to keep peace. They mean for us to use the technology to derive a new and deeper basis of understanding that escapes “human biases” and “ideological systems” (including religion) that have guided mankind from the beginning.
They quite literally do plan that the technology will lord over us in a world where, according to Elon Musk, “biological intelligence is 1%” of all future intelligence.
They intend for us to grant AI agency over ourselves in as much of life as possible, so that the machines orchestrate the majority of human activity and so we delegate decision making to AI, which they claim will be the most efficient and unbiased source that we can make decisions from.
It is a dark vision, and though it is not inevitable, adoption and use of the technology in the wrong ways could accelerate a transition to all of part of that dream.
Present Danger
There is an urgent and present danger that stems from our assumptions that this technology is truly an artificial intelligence. People are interacting with the tool as it if is a peer or even superior being. People are using it as a companion and are granting the tool agency (control) over them in much of life. In this respect, many have started using the tool to replace relationships in their life (companions) and even replace people (jobs, professions etc). In this way, AI is rapidly becoming an idol that can actually exert influence over us.
The great risk here is that we are quickly losing our being, losing our humanity, and ending our activity. We are migrating the things by which we have meaning and purpose to the machines. This can rapidly lead to an identity and meaning crisis.
Weaponization
AI has been adopted by militaries and criminals around the world and is actively being proposed to do harm. The majority of the harmful use of the technology comes in the following forms:
Human espionage to guide political, ideological, or malicious outcomes.
Mimicry of humans or other image, video, or voice-based deceptions.
Organizing and planning malicious activities meant to have the highest probability of success and the lost cost possible.
Driving military research.
Guiding military weapons and craft.
It is important to note that in the world of cybercrime, AI is still very limited and existing defenses are still effective at countering all known threats.
A Practive Approach to AI
For guidance on our recommended use of AI, see our report titled “A Practive Approach to AI.”
In summary, we recommend opting out of use of current AI technology, except in explicit cases of the designed purpose of the LLMs where the added attributes of natural language processing and generative output are used in their proper context.
Whenever using AI, remember its designed purpose, its place relative to man (as a tool), and its limitations.