I’m going to tell you something that most VCs and founders in Silicon Valley won’t. The real cost of AI safety has nothing to do with compute. It’s not about buying more GPUs or bigger models. It’s a people problem, and it’s way more expensive than you think.
I’ve seen it firsthand. At RemoteTeam, we were building AI to automate HR tasks. The pressure to ship was immense. Every investor deck was about growth, speed, and hitting the next milestone. But behind the scenes, we were grappling with a much harder problem: how to make sure our AI wasn’t biased. How to prevent it from making decisions that could unfairly impact someone’s career. That’s a cost that doesn’t show up on a balance sheet, but it’s the one that can kill your company.
The Code of Silence Around AI Safety Costs
There’s a reason nobody talks about this. It’s not sexy. It’s not the kind of thing you put in a press release. It’s the messy, complicated, human side of AI. And in a world obsessed with data and algorithms, the human element is often the first thing to get cut when budgets are tight.
Founders are afraid to bring it up with their investors. They don’t want to seem like they’re slowing down. They don’t want to be the one to say, “We need to hire a sociologist to study the impact of our product.” It sounds like a luxury, not a necessity. But I’m telling you, it’s the most important investment you can make.
The Real Financial Model for AI Safety
When I was raising money for MovieLaLa, I had a slide in my deck that broke down our costs. It had all the usual suspects: engineering, marketing, sales, G&A. But I had one line item that most investors had never seen before: “Trust & Safety.” It was 15% of our R&D budget. Some investors balked. “Can’t you just build a filter?” they’d ask. They didn’t get it.
Trust & Safety wasn’t a feature. It was a core part of the product. It was a team of people who were constantly thinking about how our technology could be misused. They were the ones who were up at night worrying about deepfakes and misinformation. They were the ones who were building the systems to protect our users.
Here’s a simplified version of the financial model I use when I’m advising startups on AI safety:
- Human Capital (70%): This is the biggest piece of the pie. It’s not just about hiring a few content moderators. It’s about building a team of experts with diverse backgrounds. You need engineers who understand fairness and bias. You need designers who can create user interfaces that are transparent and explainable. You need lawyers who can navigate the complex legal and ethical landscape. And you need a leader who can bring them all together and make sure their voices are heard.
- Tooling & Infrastructure (20%): This is where the tech comes in. It’s the monitoring systems, the explainability tools, the bias detection frameworks. It’s the infrastructure you need to build and maintain a safe and trustworthy AI system. This is the part that most people focus on, but it’s only a small piece of the puzzle.
- Contingency (10%): This is the “oh shit” fund. It’s for when things go wrong. And they will go wrong. You’ll have a PR crisis. You’ll have a security breach. You’ll have a model that goes off the rails. This fund is what allows you to respond quickly and effectively without having to go back to your investors with your tail between your legs.
It’s a Culture, Not a Checklist
I can’t stress this enough. You can’t just throw money at the problem and expect it to go away. AI safety is not a checklist you can complete. It’s a culture you have to build. It has to be in the DNA of your company.
At RemoteTeam, we had a weekly meeting called “Red Team.” The goal of the meeting was to try and break our own product. We would come up with all the ways that our AI could be misused. We would try to find the edge cases that we hadn’t thought of. It was a humbling experience. But it was also one of the most valuable things we did.
It created a culture of accountability. It made everyone in the company feel responsible for the safety of our users. It wasn’t just the job of the Trust & Safety team. It was everyone’s job.
This is the kind of cultural shift that is required to build safe AI. It’s not easy. It takes time and effort. But it’s the only way to do it right.
The Day Our Algorithm Went Rogue
I remember one Tuesday morning at MovieLaLa, our AI-powered movie recommendation engine started suggesting nothing but obscure 1950s French new wave films to our entire user base. The first reports trickled in as confused users, mostly teenagers, wondering why their feed was suddenly full of black and white movies with subtitles. Our engagement numbers plummeted. It was a five-alarm fire.
We traced the problem to a single data pipeline that had been contaminated with a bad dataset. A junior engineer, trying to be proactive, had integrated a new, unvetted source of classic film data. The algorithm, designed to find and promote hidden gems, did its job too well. It latched onto this new data and decided, in its infinite wisdom, that everyone needed a crash course in Truffaut.
Fixing the technical glitch was the easy part. We rolled back the data pipeline within an hour. The hard part was rebuilding the trust we had lost. We spent the next week personally emailing our top users, explaining what had happened, and offering them a month of our premium service for free. It cost us over $50,000 in lost revenue and customer credits. But the real cost was the hit to our reputation.
That incident was a wake-up call. It taught me that AI safety isn’t just about preventing catastrophic failures. It’s about the thousand small decisions you make every day. It’s about the data you use, the models you build, and the way you test and deploy them.
A Call to Action for Investors
As an investor in over 200 companies, including some of the biggest names in AI like Anthropic, OpenAI, and Scale AI, I’ve seen the good, the bad, and the ugly. I’ve seen founders who are deeply committed to building safe and ethical AI. And I’ve seen founders who are willing to cut corners to get to the next funding round.
Here’s my message to my fellow investors: we have a responsibility to fund AI safety. We need to stop rewarding growth at all costs. We need to start asking the tough questions about how companies are building their AI systems. We need to make it clear that we value safety as much as we value speed.
I’m not saying it’s easy. It’s a lot harder to measure the ROI of a Trust & Safety team than it is to measure user growth. But in the long run, it’s the only thing that matters. Because a company that loses the trust of its users has nothing.
This Is Not Someone Else's Problem
The conversation around AI safety is often dominated by abstract, long-term fears about superintelligence and existential risk. While those are important discussions, they distract from the immediate, tangible problems we face today. Bias in hiring algorithms, the spread of deepfake misinformation, the erosion of privacy—these are not future problems. They are happening right now.
We don't need to wait for a rogue super-AI to start taking safety seriously. The real cost of AI is not some far-off, hypothetical event. It's the daily price we pay when we prioritize speed over caution, growth over responsibility. It's the cost to individuals who are unfairly denied a job or a loan because of a biased algorithm. It's the cost to society when our information ecosystem is polluted with synthetic media that is indistinguishable from reality.
So, the next time you hear a founder or an investor bragging about their latest AI breakthrough, ask them a simple question: “What’s your budget for Trust & Safety?” Their answer will tell you everything you need to know.
The real cost of AI safety isn't a line item in a budget. It's the price of admission for building a future we actually want to live in. And it's a price we can't afford not to pay.
Three Practical Steps for Founders
Talking about culture is great, but what can you actually do on Monday morning? It’s less complicated than you think. You don’t need a PhD in ethics to get started. Here are three things I tell every founder I invest in:
Appoint a “Designated Skeptic”: In every product meeting, someone’s job should be to challenge the assumptions being made. Their role is to ask the uncomfortable questions: How could this feature be abused? What’s the worst-case scenario? What are we not thinking about? This isn’t about being negative; it’s about being rigorous. At MovieLaLa, our designated skeptic was a former journalist who was trained to think about sources and verification. She saved us from countless mistakes.
Create a “Bias Bounty” Program: Just like a bug bounty program, offer a reward to anyone—internal or external—who can find evidence of bias in your algorithms. This does two things. First, it gives you a powerful, decentralized way to audit your own systems. Second, it sends a clear signal to your team and your users that you take this issue seriously. The first time we paid out a $1,000 bounty for a bias report at RemoteTeam, it changed the conversation internally overnight.
Conduct Pre-Mortems, Not Just Post-Mortems: We’re all familiar with post-mortems—the process of figuring out what went wrong after a failure. A pre-mortem is the opposite. Before you launch a new product or feature, you get the team together and imagine that it has failed spectacularly. Then you work backward to figure out what could have caused the failure. This process forces you to confront potential risks before they become real problems. It’s a simple, powerful way to de-risk your roadmap.
These aren’t silver bullets. But they are starting points. They are practical, concrete steps you can take to start building a culture of safety and responsibility. You don’t need a massive budget to implement them. You just need the will to do it.
Frequently Asked Questions
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.