''' I’m going to tell you something that most people in Silicon Valley won’t. They’ll talk your ear off about neural networks, about scaling laws, about the latest multi-billion dollar funding round. But they won’t talk about the real cost of making AI safe. And I’m not talking about the price of a few extra GPUs. I’m talking about the messy, expensive, and absolutely critical investment in people and culture.
I’ve been in this game for a while. I built two AI companies from the ground up and sold them. RemoteTeam went to Gusto, and MovieLaLa was acquired by Gfycat. I’ve also written checks for over 200 startups, including some of the biggest names in AI today like Anthropic, OpenAI, and Scale AI. I’ve seen what works, and I’ve seen what blows up in your face. And the biggest mistake I see founders make is treating safety as a line item on a spreadsheet. It’s not. It’s the foundation of your entire company.
The Real Cost Isn't Code, It's Cognition
Everyone gets excited about the tech. The algorithms. The models. And yes, that stuff is important. But the hardest part of AI safety has nothing to do with code. It’s about people. It’s about building a team that is constantly thinking about how your technology could be misused, and creating a culture where it’s safe to raise those concerns.
At RemoteTeam, we were building AI to help manage distributed teams. We had a feature that could analyze team communication patterns to predict burnout. On the surface, it sounded great. But one of our junior engineers, a recent grad named Maria, came to me one day looking nervous. She pointed out that the same tool could be used to identify and punish employees who weren’t “team players.” She was right. We had been so focused on the positive applications that we hadn’t fully considered the dark side. That conversation changed everything. We ended up building in a number of safeguards, including making the data anonymous and only showing aggregated team-level trends. It cost us an extra quarter of development time and probably a couple hundred thousand dollars in salaries. But it was the right thing to do. And it made the product better.
This is what I mean by the human cost of AI safety. It’s not about hiring a “chief ethics officer” and calling it a day. It’s about empowering every single person on your team to think like a red teamer. It’s about creating a culture where people feel safe to challenge assumptions, even if it means slowing things down.
The Financial Models They Don't Show You
So what does this actually cost? It’s not a simple answer. But I can give you a framework for how I think about it. When I’m evaluating a new investment, I look for a few key things:
- Safety-to-engineering ratio: For every 10 engineers you have building your core product, you should have at least one person whose full-time job is to think about safety and ethics. This isn’t a hard and fast rule, but it’s a good starting point. And that person shouldn’t be a philosopher in an ivory tower. They should be a deeply technical person who can get their hands dirty in the code.
- Red teaming budget: You should be spending at least 5% of your R&D budget on internal and external red teaming. This means hiring outside experts to try and break your system, and giving your own team the time and resources to do the same. At one of my portfolio companies, we have a standing “bug bounty” for anyone who can find a way to misuse the AI. The payouts are significant. We’ve paid out over $250,000 in bounties in the last year alone.
- Training and education: You need to be constantly training your team on the latest safety research and best practices. This isn’t a one-time thing. The field is moving too fast. I recommend at least a full week of dedicated safety training for all new engineering hires, and ongoing education for the entire team.
When you add it all up, you’re looking at a significant investment. For a 50-person startup, you’re probably talking about an extra $500,000 to $1 million a year. That’s a lot of money. But it’s a lot cheaper than the alternative. A single safety incident can destroy your company’s reputation, open you up to lawsuits, and attract the wrong kind of attention from regulators. I’ve seen it happen.
Beyond the Balance Sheet
The real challenge, though, isn’t the money. It’s the culture. It’s about moving from a mindset of “move fast and break things” to “move carefully and build things that last.” It’s about rewarding people for finding problems, not just for shipping features. And it’s about having the humility to admit that you don’t have all the answers.
One of the most effective things I’ve seen is to create a dedicated “safety council” within the company. This council should be made up of people from across the organization – engineering, product, legal, and even sales. They should meet regularly to review new features, discuss potential risks, and make recommendations to the leadership team. This does two things. First, it distributes the responsibility for safety across the entire company. Second, it ensures that you’re getting a diversity of perspectives. An engineer might see a technical risk, while a salesperson might see a reputational one.
We’re at a critical moment in the history of AI. The decisions we make today will have consequences for decades to come. We have the opportunity to build a future where AI is a force for good, a tool that empowers people and helps us solve some of the world’s most pressing problems. But we can only do that if we’re willing to have an honest conversation about the real cost of safety. It’s time to stop pretending that it’s someone else’s problem. It’s our problem. And it’s time to start solving it. '''
Frequently Asked Questions
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
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.
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.