OpenAI has shut down Preparedness, the team built to catch whether its models could enable bioweapons or start hacking on their own. It's the third safety unit the company has dissolved in two years, and it lands just as OpenAI lines up a public listing that could value it above $850 billion.
OpenAI folded its Preparedness team into the rest of its research organization at the end of July 2026, according to reporting from TechCrunch. The team's job was blunt and specific: evaluate whether a new frontier model could walk someone through building a biological weapon, or whether a system might start breaching other companies' networks on its own initiative. That job no longer has a dedicated owner. Bio and cyber risk assessments have been split across existing teams instead, and nobody at OpenAI is tasked with watching the whole board. OpenAI built Preparedness in October 2023 specifically to study catastrophic scenarios, nuclear risk among them, back when the company was still fielding questions about whether it took AI safety seriously at all.
This is not an isolated reshuffle. It's the third time in two years OpenAI has dissolved a team built specifically to study catastrophic AI risk. AGI Readiness went first, in October 2024, shortly after its head, Miles Brundage, quit and wrote publicly that neither OpenAI nor any other lab was ready for AGI, as CNBC reported at the time. Mission Alignment followed in February 2026. Now it's Preparedness, four months before OpenAI is expected to go public.
Dylan Scandinaro ran Preparedness. He's reportedly shifted his own focus to risks from self-improving AI systems, according to TechCrunch's reporting on the reorganization. But the team he built to back him up is gone. He's one person now, not a unit.
The dissolutions haven't happened in a vacuum. Brad Lightcap, OpenAI's chief operating officer for four years and its chief financial officer before that, told staff on August 11, 2026 that he's leaving after eight years to start something new, according to TechCrunch. He'd already stepped back from day-to-day management in April, when chief revenue officer Denise Dresser absorbed most of his responsibilities. Chloe Bakalar, OpenAI's chief ethics officer, left the company in July 2026, less than a year after joining. OpenAI never announced it. Nobody has replaced her.
Brad Lightcap Is the Latest Senior OpenAI Executive to Head for the Exit OpenAI COO Brad Lightcap announced on August 11, 2026 that he's leaving to start a new venture, becoming the most senior exit yet in a year of departures that has also cost the company its product chief, its enterprise sales lead, and its only dedicated AI ethicist. The exodus comes just months after OpenAI closed a $122 billion round valuing it...
Fidji Simo and Barret Zoph have also exited since April. Put together, OpenAI has lost its COO, its ethics chief, and the leadership or entire structure of three risk-focused teams within roughly two years, all while telling investors the company is ready for a stock market listing.
Why the timing matters #
OpenAI filed confidentially for an IPO with the SEC on June 8, 2026, working with Goldman Sachs and Morgan Stanley, TechCrunch reported. Reporting since has pointed to a valuation between $730 billion and $850 billion, with some estimates running past $1 trillion, and a possible September 2026 debut. A confidential filing means the real numbers, revenue and margins included, stay hidden from the public for months. The safety infrastructure, unlike the revenue figures, is dissolving in plain sight.
Frankly, you don't build three separate teams to study catastrophic risk and then quietly retire all three unless something else is winning the argument internally. Speed is winning. Model releases keep shipping. IPO prep keeps moving. And the group whose entire purpose was to say wait, check this first keeps getting absorbed into teams that already have a dozen other priorities.
OpenAI says the underlying work continues, just distributed differently. Maybe it does. But distributed responsibility is a well known way to make sure nobody in particular is accountable when something goes wrong. For a company about to ask public shareholders to trust its risk management alongside its revenue growth, that is not a small detail to bury in a July reorg memo.
Also read: Microsoft Merges Its Copilot Apps as It Builds an AI Super App • Andrew Ng names the four AI skills that decide which teams actually ship • Stripe Agrees to Buy AI Router OpenRouter for More Than $7 Billion