Sam Altman just accused Anthropic of selling fear to justify keeping the most powerful AI models locked behind enterprise paywalls.
The Summary
- OpenAI CEO Sam Altman told podcaster Ashlee Vance that Anthropic is using "fear-based marketing" to sell Claude Mythos, comparing it to "building a bomb, threatening to drop it, then selling a $100 million bomb shelter."
- Anthropic is withholding public release of Claude Mythos, offering access only to select enterprise customers who can afford exclusive deployment.
- Altman argues some in tech want to keep AI "in the hands of a smaller group of people" under the guise of safety concerns, while OpenAI plans to "err on the side of larger releases."
The Signal
This is not just CEO trash talk. This is the defining fork in the road for how AI gets distributed in the agent economy. Anthropic made Claude Mythos invite-only, citing cybersecurity capabilities too dangerous for public release. Altman's response frames this as a commercial strategy disguised as responsibility.
The question is whether advanced AI becomes a luxury good or a utility. If Anthropic's model wins, the most capable agents live behind enterprise moats. Your company pays millions for access, or you're stuck with last year's model. If OpenAI's model wins, powerful agents proliferate faster, with more risk but also more distributed capability.
"There are people in the world who, for a long time, have wanted to keep AI in the hands of a smaller group of people."
Altman's phrasing matters here. He's not just defending OpenAI's release strategy. He's accusing competitors of using safety rhetoric as cover for market capture. The implication: if you control who gets the most advanced models, you control which companies can build competitive AI agents. You become the gatekeeper to Web4.
The irony is thick. OpenAI started as a nonprofit explicitly to prevent AI concentration. Now Altman positions OpenAI as the open alternative to Anthropic's walled garden. Meanwhile, OpenAI has its own enterprise tiers and hasn't exactly been giving GPT-5 away for free.
Here's what's actually at stake:
- Agent capability distribution: If only Fortune 500 companies can afford Mythos-class models, small companies can't build competitive autonomous agents
- Security theater vs. real risk: Is withholding Mythos genuinely protective, or does wide release actually make systems more secure through distributed testing?
- The safety narrative as moat: Companies that claim their models are "too dangerous" for public use also create artificial scarcity and premium pricing power
Anthropic's model creates a two-tier system. Tier one: companies that can pay for exclusive model access and build agents that outperform everyone else. Tier two: everyone else, building on older, weaker models, falling further behind. That gap compounds. The company with better agents wins customers, affords better models, builds better agents.
Altman knows this. His bet is that OpenAI wins by releasing powerful models to more developers, who build more agents, which creates more demand, which funds better models. Network effects favor the platform with the most builders. But that only works if OpenAI's models stay competitive with Anthropic's best locked offerings.
The Implication
Watch which companies get Mythos access. That list tells you who Anthropic thinks will pay premium rates and who they're willing to make more capable than their competitors. If you're building AI products and not on that list, you're already behind. The agent economy splits into haves and have-nots based on model access, not just talent or capital.
For developers, Altman's framing is convenient but not wrong. Broader model access does benefit builders. But "benefit everybody" from a company valued at $100+ billion still means benefit through their API, on their terms. The real question is whether any company, OpenAI or Anthropic, should decide who gets to build with the most capable intelligence. That's a Web3 problem dressed up as a Web4 fight.