Three companies that control most of the world’s frontier AI have been meeting in private for weeks. OpenAI, Anthropic, and Google are calling it AI safety. I call it the biggest regulatory capture play since the banks rewrote their own rules after 2008. The companies with a seat at this table are building the walls that keep everyone else out.
What Is Actually Going On
For weeks, executives from OpenAI, Anthropic, and Google have been in closed door talks about AI safety standards and governance. According to reporting from multiple major news outlets, the conversations center on testing protocols, model evaluations, and shared safety benchmarks.
On the surface, this sounds responsible. Three powerful companies working together to make AI safer. Good for humanity, right?
But look at who is not in the room. Meta is not there. Mistral is not there. The dozens of open source labs pushing frontier models are not there. The startups trying to compete with the big players are not there.
According to the Stanford AI Index Report, the number of foundation model releases from industry reached 51 in 2023, up from just a handful two years prior. That level of competition is exactly what this “safety” framework is designed to manage. And not in the public’s favor.
The Real Play Here
I’ve watched this pattern before. Industry leaders get together, say they’re protecting consumers, then write rules that only they can afford to follow.
The banking sector did it after 2008. Pharmaceutical companies do it with FDA compliance costs every decade. Now Big Tech is doing it with AI.
OpenAI was valued at $157 billion in its most recent fundraising round, according to Bloomberg. Anthropic has raised more than $7 billion, with Amazon committing $4 billion alone, according to the company’s own announcements. Google has invested billions more in both its own models and in Anthropic as a hedge.
These three companies have enough capital to absorb any compliance cost. A startup trying to build a competing model does not.
So when the safety framework requires third party audits that cost $500,000 per model evaluation, or mandatory federal registration before deployment, or six-month prelaunch review periods, the big players shrug. The small players shut down.
That is not safety. That is a moat with a friendly name.
The contrarian read: the companies most likely to cause harm are the ones writing the rules to prevent it. And the companies most likely to build something better, faster, and cheaper are being locked out before they get started.
For business owners watching this, there is a practical lesson here. The AI tools you use right now, from customer service to financial modeling, may soon be priced more by cartel agreements than by real competition. When three companies control the frontier, they control the price. And the price goes up.
Tracking what you spend on these platforms matters more now than it did two years ago. A business card platform like Wallester lets you set spending controls and issue virtual cards for each AI subscription so you can see exactly where your money is going as prices shift month to month.
What This Means for You
If you run a business that relies on AI tools, here is what I would do right now.
First, diversify your AI stack. Do not rely on a single provider. OpenAI, Anthropic, and Google all want you locked in. The moment a standards framework passes that favors these three, the switching costs go up. Use multiple providers now while competition still keeps pricing honest.
Second, pay attention to open source alternatives. Models from Meta’s LLaMA family and projects from European and independent labs are not subject to whatever private framework these three companies agree on. If your use case fits a smaller model, run it yourself. Own the compute. Own the output.
Third, hire smart now. The AI safety conversation is creating new roles: compliance officers, red team testers, model evaluators. Companies hiring in AI right now are going to face new regulatory requirements within 12 to 18 months. If you’re building a team, use a payroll tool like Gusto to track your AI-related hires separately so you can see the real cost of compliance as it scales with your headcount.
According to McKinsey’s 2024 State of AI report, 65% of companies say they’re regularly using generative AI in at least one business function. That number keeps climbing. The question is not whether AI costs you money. The question is whether you see the bill coming before it hits.
The companies that win the next five years are not the ones using the most AI. They’re the ones using it efficiently enough to survive when pricing power consolidates at the top.
The Bottom Line
OpenAI, Anthropic, and Google are not meeting to save the world. They’re meeting to decide who gets to charge you for it. The safety framework that comes out of these talks will look reasonable, probably pass without serious opposition, and then the walls go up. Diversify your AI vendors, track your spend, and get your infrastructure in order before the drawbridge closes.
Frequently Asked Questions
What are OpenAI, Anthropic, and Google discussing in their AI safety talks?
The talks center on shared safety standards, model evaluation protocols, and governance frameworks for frontier AI systems. All three companies have framed this as a voluntary effort to align on responsible deployment before government regulation forces the issue.
Is this AI safety framework legally binding?
Not yet. These are private talks between companies. But frameworks like this often become the foundation for future legislation, and the companies writing the standards today are positioning themselves to influence how those laws get written.
How do the AI safety talks affect small businesses using AI tools?
In the short term, not much. In the medium term, reduced competition and compliance costs could push prices up for AI APIs and platforms. Diversifying your vendor relationships now reduces your exposure if one provider raises prices or changes terms after a framework locks in.
Why are Meta and open source labs not part of the AI safety talks?
Meta has taken a different path by releasing open source models, which makes centralized governance harder to apply. Open source labs also lack the lobbying resources and institutional relationships that put OpenAI, Anthropic, and Google in closed door conversations with policymakers in the first place.
Should I be worried about AI safety as a business owner?
The risks most relevant to business owners are not existential scenarios. They’re pricing power, vendor lock-in, and regulatory compliance costs. Those are the AI safety issues worth tracking right now, not the science fiction version.
