Critical Path #1 - Why everyone lies about AI
In a Nutshell
**Shadow AI is already everywhere—40-80% of employees use unapproved generative AI tools regardless of policy—and companies must stop fighting it and instead implement amnesty, anonymous surveys, clear governance, and safe on-ramps to bring it into the open.** The core mistake is treating AI as an IT issue to delegate; CEOs and boards must personally gain fluency with frontier models to lead adoption safely and effectively. The five-step framework—amnesty, surveys, governance, on-ramps, and onboarding—converts risky shadow usage into controlled, value-creating "sunlighted AI" that amplifies human intelligence without creating data breaches or hallucinations.
These notes were generated by AI and may contain inaccuracies.
Organizations cannot fight the adoption of AI tools. Employees will use AI regardless of policies, with some taking photos of work laptop screens on personal phones to bypass firewall restrictions when chatbots are banned. This reality exists across every laptop and every firewall, whether officially approved or not.
Critical Path is a podcast focused on what executives and leaders can do across every industry in the face of generative AI. The podcast is a relaunch with a new direction. David Shapiro has been in technology for almost 20 years, spending the first 15 years in IT infrastructure, cloud automation, and the last five focusing on artificial intelligence, specifically generative AI.
Dalor Petravvic is senior partner at Deloitte (retired) with more than 30 years of industry experience in business transformation through technology and leadership activation. His personal mission is to help business owners and business leaders successfully navigate their organizations through the transition into the age of intelligence and age of automation.
The mission is to provide clarity and confidence to business leaders on how to think about this moment, what intelligence and automation mean personally and professionally, and to provide guidance on how to navigate, organize, and mobilize organizations into their future versions. The goal is to create organizations that are more intelligent, more automated, and more autonomous, including sovereignty considerations.
The hosts believe this represents the single most transformative period in human history, including business. David's personal mission involves building the automated future while recognizing there will be significant work to transition to post-labor economics.
Shadow AI is borrowed from the concept of shadow IT, referring to dimensions of technology that organizations are unaware of. This includes unapproved software and optional policies. AI, specifically generative AI in the form of chatbots and now agents, forms the next layer of shadow IT.
There is a proliferation of employees using AI, both approved and unapproved. This represents both a risk that needs to be understood and an opportunity that needs to be taken advantage of. Organizations are struggling with how to properly corral this activity, not block or prevent it, but organize to deliver it safely and productively.
Statistics show 40% to 80% of employees are using AI whether approved or not. The hosts note caution with these statistics as some surveys may not be robust or rigorous, but front-line observations from CTOs, CEOs, VPs, directors, and individual contributors confirm widespread usage at every level.
The C-suite is actually the biggest transgressors of using unapproved AI because they feel they know what they're doing and have the prerogative to use ChatGPT, Claude, Gemini, or whatever they choose. Middle management uses AI to keep up because companies are cutting middle management layers. Individual contributors use generative AI to get ahead of peers or safeguard their positions.
Many people lie about usage. Surveys indicate 26% to 46% of individual contributors will use unapproved AI even when outright banned. The incentive structure drives this behavior because everyone needs to keep up but doesn't want to get punished due to the risks involved.
You cannot fight this. Invite it. Invite it properly.
Intelligence is now available to people whether approved or not. The message is that you actually want your people to have 20 IQ points more every day on tap as required. The question should not be how to control shadow AI, but how to corral this activity, surface it without penalty, and understand who should have access to frontier models that are safe, protected, orchestrated, and managed within the enterprise.
AI transformation should be compartmentalized into two linked but separate activities:
- Intelligence deployment: Deploying frontier models to people to enable them to be more productive, more creative, more intelligent
- Automation: Leveraging technology to automate processes, remove bottlenecks, remove manual work and software
Many people confuse these two and refer to both as AI. Automation can be intelligent but doesn't have to be. Many automation experts argue against having generative AI on automation systems because you want predictable results.
The most common posture is the "don't ask don't tell" policy with generative AI. Executives recognize they're using it, their VPs are using it, and legal staff are using it, but they ignore it until something blows up. Data breaches cost an average of $670,000. Hallucinations present another major risk, with Deloitte having to refund an entire client fee due to hallucinations in a report, and law firms and airlines facing lawsuits over hallucinations.
Other problematic postures include:
- Blanket bans: Willfully blind to what's actually happening since usage continues regardless
- Forced quotas: Treating AI as a first-class priority with required usage quotas, which incentivizes bad behavior as people use tools whether or not they're good for the task
Artificial intelligence is far too important as a general technology to be delegated by the CEO to an IT person to lead. True leaders need enough curiosity to personally understand what AI actually is, its power, potential, and limitations.
The first step for any organization is for the chair of the board, directors, and CEOs to raise their own fluency around frontier models. This requires a personal mindset shift rather than formal training. Once executives become fluent, they can determine who in the organization ought to be empowered by this technology in an orderly fashion cascading down.
The balanced response involves approved usage that is safely constructed and given for business-rationalized purposes. Large organizations that extended safe access like Copilot to everybody saw prices go very high with value not matching the cost.
- Amnesty: Create amnesty policies so people understand they won't be punished for speaking up about their AI usage
- Surveys: Use surveys to gain situational awareness, ensuring they are genuinely anonymous rather than pseudo-anonymous
- Governance policy: Provide guidance that middle management can point to regarding approved and disapproved uses, and how to get new use cases approved
- On-ramps: Create pathways for new tools, processes, or use cases to be approved rather than blanket rejections
- Onboarding: Send surveys, provide training on governance, and implement policies like "bring your own AI" with clear guidelines about what can and cannot be put into personal AI systems
Internal use policies should not be overcomplicated. Medium businesses need a one-pager describing what is allowed and what is not allowed. The goal is to get people to honestly tell what they're using and for what reason, which will reveal ideas for improving the business.
The approach starts with executives using their personal home computers and phones because access to frontier models is typically not approved inside organizations. There is no time to wait for enterprise versions to be approved. Executives need to be introduced to these capabilities safely and quickly.
Safely configure their own personal device to set up their personal projects and stand them up as their chief of staff, accessing only publicly available documents that are relevant and helping them configure that project properly. This is the beginning of the proper experience for a CEO and executive, feeling what additional intelligence can actually do when it's properly set up and configured at the get-go.
This is something that is going to shape and define your legacy. The next five years is going to be shaped and defined by how you have navigated yourself and your organization through this. This is general technology. It's like electricity, like water, like fire. It's going to change everything. You owe it to yourself to first and foremost overcome the hesitation, overcome potential anxiety about you might not being a technologist yourself. You don't need to be a technologist to use this. You have to cross the bridge and start using it.
When the owner starts using this, when the owner has seen what the potential is, they will champion it to their next layer of leaders and that propagation is going to start occurring and cascading down. That's the right path. When you are on that path, then very quickly you will come to the realization: "Oh my God, many people in my company are using this now. I know why they are using this now. I can tell why because I'm personally getting the benefits now."
Come together and introduce these five steps to provide people with comfort: the policy about amnesty, the definition of approved usage, actually agreeing what's approved, and how are we going to access it. Those choices do matter both for companies inside the United States and organizations outside of the United States. Data residency, sovereignty, all of these factors need to be considered, but they can be solved. We now know how to solve those.
You will be very well served if you were to directly address and help your shadow AI become sunlighted AI, well illuminated, and then properly balanced to the degree where it financially makes sense. While eventually financially it will make sense to go all the way in and that every human that is employed will eventually find a way to be empowered and strengthened by AI, that might not be the case at this moment in time. So you need to be mindful about how deep do we go. Illuminating shadow AI is going to be a very important milestone along the way.
The general emotional arc that executives and business leaders go through repeatedly: they start with skepticism, often by the time they get to us they might not be skeptical anymore but they often started skeptical. Then there's excitement: "Wow, this is cool, I use this all day every day." Then there is often fear or anxiety as they realize: "Wow, this is big. This is going to change everything." Followed by paranoia as they realize that everyone is using it and that they don't know what they don't know. Then finally, after the blackening of all the emotions and all the fear and the grief, then the excitement starts to come back. That's usually when they're working with us because they're like, "These guys know their shit."
Dalibore has been at the consulting gig for longer than David. The Summit Leadership Alliance is an elite expert network of people who are like us, senior leaders who are curious and intensely focused on solving this exact problem: How do we help business owners and business leaders quickly navigate confidently navigate safely navigate themselves and their organizations to this transition. Dalibore has made this his personal mission.
Those who follow David's work around post-labor economics and L0 would recognize that David has identified the moment we are in as truly one of the most important critical moments. How we collectively navigate ourselves through this next 5 years, next 10 years will determine whether we end up in an age of sustainable abundance for all or whether things go really south on us. Both David and Dalibore and many others are intensely focused on contributing what they know and what they can to help navigate us towards that better future.
David has already discussed and solved and presented his point of view on a societal impact of this in his magnum opus L0 book, and in that theory we address the societal and governance, government interventions, as well as how individuals and families need to think about what is coming and how they should be reacting. Now we have the middle piece which is the industry, the piece that connects individuals and families and society broader. Now we're going to be spending time, energy and focus to help address that middle part.
The white paper that David has prepared for all the research that we've done for Shadow AI is available to Patreon and Substack subscribers right now. If you get in touch with Dalibore from one of his platforms, David will make sure that he has access to give that to you as well.
This podcast will be going out on multiple platforms: Dave Shapp's main channel on YouTube, Dalibore's YouTube channel Tectonic Conversations, as well as a Substack by the same name Tectonic Conversations. This will also go out on Dave's Substack. Periodic updates will be going out mostly on Substack. There'll be email address to get in touch for consulting inquiries. Like and subscribe. We will see you all next time.
Keep David Shapiro in your library
Save the videos and channels worth coming back to, and find them again in one place.
