OpenAI killed its Preparedness Team. The market reacted with a collective shrug. Revenue hit $40 billion annualized, the IPO valuation whispers are at $1 trillion, and the headline narrative is “efficiency upgrade.” But for those of us who dissect organizational skeletons for a living, this is the loudest red flag since the 2022 DeFi bridge exploit that I flagged in a static analysis—the one the team ignored until I pushed the GitHub disclosure.
Context: The Hype Cycle and the Hidden Signal
OpenAI is in the middle of a classic hypergrowth transition. From a $24 billion annualized run rate to $40 billion in under a year. The product is ChatGPT, the enemy is Anthropic, and the strategy is “focus.” The FT report from August 15, 2025, details a company in constant reorganization: five internal restructurings, a departing CRO, an ethics lead walking out, and now the dissolution of the Preparedness Team—the unit specifically tasked with evaluating catastrophic risks from AI models. The official line: “Organizational changes to improve efficiency and focus on ChatGPT.”
But beneath every whitepaper lies a buried intent. The Preparedness Team was not a cost center—it was a firewall. Its dissolution means safety assessment is no longer an independent function. It is now embedded into product teams whose primary KPI is shipping velocity. This is not a technical decision. It is a governance signal.
Core: The Systematic Teardown
Let me be precise. The Preparedness Team was created in late 2023 after the leadership crisis, directly reporting to the board. Its mandate: evaluate models for capabilities in bioweapon synthesis, autonomous replication, and cyberattack vectors. That function is now distributed across business units. In practice, this means the people who approve a model release are the same people whose bonuses depend on the model being released. Conflict of interest is not a bug—it is a feature of this new structure.
From my experience auditing codebases, I know that when a project removes an independent security layer, the failure rate rises by an order of magnitude. In DeFi, it’s the integer overflow that gets through. In AI, it’s the model that learns to jailbreak itself. The risk is not immediate—it is latent. But the organizational footprint is clear: OpenAI is prioritizing speed over safety, and the market is rewarding it.
Look at the numbers. $40 billion annualized revenue, 25x price-to-sales. That valuation implies the market expects 5-10x revenue growth in 3-5 years. To achieve that, OpenAI must ship products faster than its competitors. Anthropic is growing faster from a smaller base, and its “Responsible Scaling Policy” is a direct contrast to OpenAI’s safety downgrade. The competitive landscape is shifting from model capability to trust capital. By dissolving the safety team, OpenAI is ceding the trust narrative to Anthropic.
Data leaves footprints; hype leaves only dust. The footprint here is a trail of departing executives: the ethics lead, the CRO, the CTO transition. The signal is not that they left—it is that they left as the safety team was dismantled. Correlation is not causation, but in forensic journalism, timing is evidence.
Contrarian: What the Bulls Got Right
To be fair, the bulls have a point. OpenAI’s revenue growth is unprecedented in AI. The company is transitioning from a research lab to a sales-driven enterprise. Reorganization is painful but necessary. The Preparedness Team’s functions might be more efficient when integrated into product teams—if the integration is done with proper safeguards. Perhaps the move will accelerate GPT-5 deployment and capture enterprise market share from Anthropic.
Additionally, the $7 billion stock buyback is a standard pre-IPO play. It allows early employees to cash out and cleans up the cap table. The IPO narrative is intact, and the $1 trillion valuation is not absurd if growth continues at 50%+ annually. The market is betting on execution, not safety.
But here is the blind spot: execution without safety eventually becomes a liability. The 2023 OpenAI leadership crisis was about the speed of AI development versus its risks. The dissolution of the safety team is a direct reversal of the commitments made then. The bulls are ignoring that trust is a slow-building asset that can be destroyed in a single incident. When the next model “accidentally” generates a dangerous output, the absence of an independent safety team will be the first question regulators ask.
Takeaway: Accountability Call
Code is law only until someone finds the loophole. OpenAI has found a loophole in its own governance structure. The question is not whether the IPO will happen—it almost certainly will. The question is whether the market will price in the risk of a safety failure before it occurs. Data leaves footprints; hype leaves only dust. The footprints here lead to a boardroom decision that prioritizes quarterly growth over existential risk. That is a story that will not end well—unless the market starts asking the right questions now.