Another safety team—gone.

On August 16, the Financial Times revealed: OpenAI disbanded its Preparedness Team at the end of July. The team's core mandate was evaluating whether frontier models posed catastrophic risks—biosecurity, cyberattacks, autonomous replication capability—and designing mitigation strategies.

After the disbanding, its responsibilities were split by domain—biosecurity, cybersecurity—and merged into existing business teams. Without an independent safety assessment group, risk judgment and product advancement now sit in the same hands.

This isn't the first time. It is the third safety team OpenAI has dismantled in three years.

Three Safety Teams, All Gone in Three Years

Let's count them:

In 2024, the Superalignment team became effectively defunct. Co-lead Jan Leike left with a public statement that the company's emphasis on safety was being replaced by product development. The team once tasked with "solving superintelligence alignment" was gradually scattered, reorganized, and marginalized.

In early 2026, the AGI Readiness team was dissolved. This team assessed how close the company was to AGI and what preparations were needed. After it was gone, no one systematically tracked or publicly disclosed AGI timelines and benchmarks.

In July 2026, the Preparedness team was disbanded. This team was the final safety gate before model releases—if it found a model posed significant risk, it had the authority to stop the launch. Now the gate is gone, and safety assessment has been "pushed down" to individual business lines.

The official line every time is roughly the same: it's not that safety matters less, it's that embedding safety into business lines is more efficient. But the simple logic is: when a safety team is independent, it has the power to say "no" to products. When safety sits inside business lines, it reports to product teams—asking subordinates to block their boss's decisions rarely works well.

Meanwhile, the executive exodus continues. This year alone, at least 13 core executives have left—including the CFO, COO, ethics lead, and chief revenue officer—most with tenures under a year. This isn't normal talent flow; it's a signal of severe organizational upheaval.

The Safety Cost on the Eve of IPO

What is all this happening against? OpenAI is preparing for an IPO. Valued at $852 billion, it has confidentially filed an S-1 with the SEC in what could be one of the largest tech IPOs in history.

What does IPO have to do with safety? A lot. Capital markets reward growth, efficiency, predictable product cadence. What is an independent safety team? A speed bump. A brake. An uncertainty—you never know if it will step in before some big product launch and say "no, there's risk."

Dismantling the safety team and pushing responsibilities down to business lines looks like organizational optimization on the surface. In practice, it moves the "can we launch?" decision from safety teams into product teams' hands. What are product teams accountable for? Growth. Revenue. KPIs.

This isn't unique to OpenAI. The entire AI industry faces the same tension: the faster you run, the harder safety is to keep up with; the more commercial you become, the more safety gets marginalized. The difference is just that some companies are still holding on, while others have quietly removed the brakes.

What does this mean for users and enterprise customers? It means you can no longer take for granted that "OpenAI has robust safety measures." When a company systematically contracts safety governance ahead of its IPO, you'd better add extra safeguards of your own—independent security audits, strict access controls, human content review. These can't be skipped.

See you tomorrow.