I rewatched the original Jurassic Park the other night. It holds up, by the way. The practical effects are still better than half of what gets greenlit today, and the pacing is tighter than most modern blockbusters that run forty minutes too long. But this time around, the thing that stuck with me wasn't the T-Rex breakout or the kitchen scene with the raptors. It was John Hammond.
Specifically, it was the way he kept saying "we spared no expense" while standing in the middle of a park that was falling apart because he cheaped out on everything that actually mattered.
Think about it. The man cloned dinosaurs. He built a theme park on a remote island off Costa Rica. He had a fleet of custom Ford Explorers running on electric rails through paddock tours. He had an animatronic flea circus in the visitor center, for crying out loud. The dining room served Chilean sea bass. The presentation was absolutely world-class. If you were an investor sitting in that helicopter on the way to Isla Nublar, watching the waterfalls and the lush greenery roll by, you would have been completely sold. This guy clearly spared no expense.
But the IT infrastructure? One guy. Dennis Nedry. One underpaid, overworked, resentful systems administrator running the entire technology backbone of a facility housing the most dangerous animals that have ever existed. No redundancy. No separation of duties. No backup operator. No change management process. The guy literally had the keys to every system, every door lock, every fence, every security camera, and every piece of critical infrastructure on the island. He was the architect, the operator, the help desk, and the on-call rotation, all rolled into one increasingly unhappy person.
And Hammond's response to Nedry's complaints about compensation and workload was basically "you should have negotiated a better contract." That's it. That was the risk mitigation strategy for the single human being who could shut down every safety system on an island full of apex predators. Negotiate better next time.
I've been thinking about this a lot lately because I keep seeing the same pattern play out with AI.
Organizations are rushing to deploy AI with the same energy Hammond brought to growing velociraptors. The demos are incredible. The pitch decks are immaculate. The executive presentations practically sparkle. The board gets a live walkthrough of the chatbot answering customer questions, or the model summarizing contracts in seconds, or the copilot writing code that looks clean and compiles on the first try. Everyone in the room is impressed. We spared no expense.
Then you look at how it's actually being secured and governed, and you find a skeleton crew, duct tape, and a prayer.
A friend of mine told me about an AI initiative at his company with a seven-figure budget where the security review was... an email thread. Maybe a spreadsheet someone started three months ago that nobody had updated since. The data governance policy was "we'll figure that out later." The access controls were "whoever has the API key." The model inventory was "ask Dave, he might know." The risk assessment was a two-page document that someone copied from a template they found online, changed the company name, and called it done.
Sound familiar? It should. That's Nedry's desk. That's one person (or one small team) being handed the keys to everything and told to make it work with what they've got while the real budget goes to making the dinosaurs look amazing.
Here's what Hammond got wrong, and what I see getting repeated almost everywhere right now. He thought the impressive parts were the important parts. The dinosaurs. The visitor experience. The wow factor. The fences were an afterthought. The software was a cost center. The guy running it all was a line item to be minimized. Hammond could describe every species in the park, knew their feeding schedules, knew which ones were his favorites. But I doubt he could have told you what operating system the security infrastructure ran on, or what would happen if the primary network went down, or whether there was a failsafe that didn't depend on a single disgruntled employee being in a good mood.
With AI, I see the same thing. The model is the dinosaur. Everyone's excited about the model. What can it do? How fast does it generate output? Can it write code? Can it summarize documents? Can it talk to customers? Can it analyze a quarter's worth of financial data in thirty seconds? Executives can tell you the name of the model, the vendor, the use case, and the projected ROI. That's the Brachiosaurus standing in the field while the music swells and everyone's jaw drops.
But the security architecture around it? The data classification? The prompt injection testing? The monitoring of what the model is actually doing with sensitive data? The access controls? The audit logging? The supply chain risk of the model itself and the training data that went into it? Those are the fences. Just like Hammond's fences, they're only as good as the infrastructure running them and the people maintaining them.
I'll be honest. I've been guilty of this myself. Early in my career, I got excited about new capabilities in other emerging technologies and treated the security and governance around them as a follow-on exercise. Something to bolt on once the cool stuff was working. Get it deployed, show the value, then we'll circle back and lock it down. I've said those words (pre my entry into security when I was still IT Operations). I've been in rooms where I nodded along when someone else said them. I learned (the hard way, more than once) that bolt-on security is how you get a Nedry situation. It's how you end up with a single point of failure that has God-mode access, no oversight, and a growing list of reasons to be unhappy about it.
The thing about the Jurassic Park disaster that I think gets overlooked is how many warning signs were just... ignored. The movie doesn't hide them. The park wasn't even open to the public yet and they'd already had a worker killed by a raptor in the opening scene. Hammond's own investors were nervous enough to demand an outside review, which is how Grant, Sattler, and Malcolm ended up on the island in the first place. The lawyer, Gennaro, was there specifically because people with money on the line were worried the park wasn't safe. The amber was cracking before the fences went down.
I see the same thing with AI deployments. The warning signs are there. Security teams are raising flags about data handling and getting told "we'll address that in phase two." Compliance is asking about regulatory requirements and getting back a vague slide about "responsible AI principles" that doesn't actually commit to anything specific. Someone in the SOC noticed the AI platform is making API calls to services nobody approved, and the ticket got deprioritized because it wasn't a production outage. The canary is singing, but the people with the budget are watching the dinosaurs.
There's a scene in the movie where Hammond is giving the tour and the dinosaurs don't show up on schedule. The Dilophosaurus is a no-show. The T-Rex won't come out to eat the goat. Hammond is visibly frustrated because his carefully choreographed demonstration isn't going according to plan. He can't control the dinosaurs. He spent all this money creating them, built this entire park around them, and they won't perform on cue.
That's what happens when you build something powerful and assume it will behave the way your demo script says it will. AI models don't perform on cue either. They hallucinate. They make confident assertions about things that aren't true. They leak training data. They follow prompt injections embedded in documents they're asked to summarize. They do things their creators didn't anticipate because the creators were focused on what the model could do, not on the full range of what it might do when it's loose in a production environment with real data and real users doing unpredictable things.
Hammond never planned for what happens when the dinosaurs don't follow the script. Most AI deployments I've seen don't either.
This is the part that bugs me the most on the rewatch. The movie wants you to blame Nedry. He's the villain. He turned off the fences, he stole the embryos, he got what was coming to him courtesy of that charming Dilophosaurus. The audience is supposed to think "what a scumbag" and move on.
But Nedry wasn't the root cause. Nedry was the predictable outcome of a system designed to fail. One person with total access, no meaningful oversight, no segregation of duties, underpaid for the criticality of their role, and bitter about it. If you ran a tabletop exercise on InGen's IT organization (and wouldn't that have been something), the insider threat scenario writes itself. It's the first thing any halfway competent risk assessment would flag. Single administrator with unrestricted access to all critical systems, no secondary approval required for major changes, no monitoring on privileged activity, and an active compensation dispute with management. That's not a risk you accept. That's a risk that should keep you up at night.
If it wasn't Nedry selling embryos to Dodgson, it would have been a bad config change at 2 AM when he was tired and careless. Or a phishing email he clicked because he was juggling too many things. Or just plain burnout leading to a mistake nobody caught because nobody else understood the systems well enough to notice. The espionage angle makes for a better movie, but the disaster was baked in long before Nedry met Dodgson at that diner. Hammond built a system where one unhappy person could bring the whole thing down, and then he made that person unhappy. That's not a technology failure. That's a leadership failure.
When I look at how some organizations are deploying AI right now, I see a lot of Nedrys waiting to happen. Not because the people are malicious, but because the structure is wrong. One team owns the AI platform with minimal security involvement. The data science group has broad access to production data with no DLP controls. The API keys are shared in a Slack channel (I wish I were making this up, but I've seen it more than once). The model can access sensitive customer data and there's no monitoring on what it returns or who's asking. The person who set up the AI infrastructure six months ago has since moved to a different team and nobody's reviewed their access. The documentation is whatever's in their head, and they're not around to ask anymore.
You don't need corporate espionage to have a disaster. You just need a system with no guardrails and enough complexity that nobody fully understands what it's doing. That's Jurassic Park. That's also a lot of AI programs I've seen in the wild.
There's another layer to this that I keep coming back to, which is the way Hammond reacted when things started going wrong. He didn't pivot to crisis management. He didn't empower his team to make hard decisions. He sat in the dining room eating ice cream and talking about how the park was going to work. He was still selling the vision while people were being eaten. His instinct, even in the middle of a catastrophe, was to defend the investment rather than face what was actually happening.
I've watched executives do this with security incidents. Not literally eat ice cream (although one time, close enough), but the same basic behavior. The instinct to minimize, to protect the narrative, to keep talking about the roadmap while the SOC is on fire. "It's a minor issue, we're handling it." "This doesn't change our strategy." "We're still on track for the Q3 launch." The dinosaurs are out and eating people, and leadership is still talking about visitor projections.
When your AI model leaks customer PII because nobody configured the output filtering, and the first response from the program owner is "this is an edge case, it won't happen again," that's Hammond eating ice cream. When your chatbot gets prompt-injected into saying something that ends up on social media and the executive sponsor's reaction is to blame the user for being adversarial, that's Hammond eating ice cream. The system failed because it wasn't built to handle failure, and the people in charge are defending the architecture instead of fixing it.
Hammond's mistake wasn't ambition. Cloning dinosaurs was genuinely impressive (setting aside the "your scientists were so preoccupied with whether they could that they didn't stop to think if they should" speech, which, fair). His mistake was thinking that spending money on the visible, exciting parts was the same as building something that wouldn't kill people. He confused spectacle with resilience. He thought that because the park looked finished, it was finished. The visitor center was beautiful. The brochures were printed. The merchandise was ready. Everything the public (or the investors) would see was polished to a mirror shine. Everything behind the curtain was held together with one guy and a prayer.
If you're deploying AI in your organization right now, and I assume most of you are in some form, ask yourself where the money is actually going. Is it going to the model, the use case, the vendor demo, and the executive update? Or is some meaningful portion of it going to the boring stuff? The access controls. The data governance. The monitoring. The incident response plan for when the model does something unexpected. The red teaming. The prompt injection testing. The model inventory so you actually know what's running where. The training for the people who are supposed to manage this stuff. The headcount so it's not one person's side project. (Especially the headcount.)
Because "spare no expense" means nothing if the fences run on one guy's laptop.
I think about the Malcolm character a lot in these conversations. Not the chaos theory monologue (although that's fun at parties) but the dinner scene. Malcolm is sitting at that table, telling Hammond to his face that the park is going to fail. Not might fail. Will fail. He's explaining exactly why, in clear terms, with a logical framework. Hammond just... smiles. Nods. Offers more food. Doesn't engage with the substance at all. He's a polite, grandfatherly man who has decided the concerns don't apply to him because he believes in what he's building.
I've sat in meetings that felt exactly like that. Security raises a concern about the AI deployment. We lay out the risks. We explain what could go wrong and what it would cost. We're not being dramatic about it. We're being specific. Leadership nods politely, thanks us for the input, and moves on to the next slide in the roadmap. The concern gets logged somewhere. Maybe it becomes a bullet point in a risk register that nobody reads. Maybe it gets assigned to someone who doesn't have the authority or the budget to do anything about it. The park opens on schedule.
The uncomfortable truth is that Hammond isn't a villain in the traditional sense. He genuinely believed he was building something wonderful. He loved those dinosaurs. He wanted to share them with the world. His intentions were good. None of that mattered when the systems failed because good intentions don't stop a T-Rex from walking through a broken fence.
I think a lot of the people driving AI initiatives right now have good intentions too. They're genuinely excited about the technology. They believe it will help their organizations and their customers. They're not trying to build something dangerous. But excitement without engineering discipline is how you get Jurassic Park. It's how you get a system that works perfectly in the demo and collapses the first time something unexpected happens. In security, unexpected things don't just happen. They're the whole job.
The dinosaurs got out because the system was designed to look impressive, not to be resilient. If that doesn't describe at least one AI deployment you've seen in the last twelve months, you're either very lucky or you're not looking hard enough.
Are you building a park, or are you building a system that works when things go wrong?