⏰ 6 min read | August 17, 2026
The OpenAI Preparedness team — the group whose entire job was to spot dangerous AI risks before they became real-world problems — has quietly been shut down, according to a new report from the Financial Times. If you have been following the drama around AI safety teams shrinking across the industry, this one stings a bit more, because Preparedness was supposed to be the group standing between a powerful new model and a genuinely bad outcome. Let’s break down what actually happened, why it matters, and what it says about where OpenAI’s priorities might be heading.
What Happened: OpenAI Disbands Its Preparedness Team
Here’s the short version. At the end of July 2026, OpenAI quietly folded its Preparedness team into other parts of the company. This wasn’t a case of the team being fired outright — instead, its members were reassigned to existing teams, with some now working on very specific slices of risk like cybersecurity and biosecurity, while others moved on to entirely different research questions. The news came to light through Financial Times reporting rather than an official OpenAI announcement, which is itself a little telling. Big structural changes like this usually get buried in a blog post update or a line in someone’s LinkedIn bio, not a press release.
Did You Know? OpenAI originally created the Preparedness team back when the company was under heavy pressure to show regulators and the public that it took catastrophic AI risks seriously — think along the lines of AI-assisted cyberattacks or bioweapon design, not just chatbots giving wrong answers.
Why This Team Mattered So Much
To understand why people are raising eyebrows, you need to know what the Preparedness team actually did. Their job was to look at OpenAI’s most advanced models and ask uncomfortable questions: could this model help someone build a cyberweapon? Could it be misused to design something dangerous in biology or chemistry? And if the answer was "maybe," their next job was to figure out how to reduce that risk before the model ever reached the public.
This kind of work sits apart from the usual AI safety chatter about bias or hallucinations. It’s about worst-case scenarios — the stuff that keeps AI policy folks up at night. Having a dedicated team meant this risk assessment wasn’t just a side task squeezed in between product launches; it was someone’s whole job to think about what could go catastrophically wrong. Folding that team into broader groups doesn’t necessarily mean the work stops, but it does mean the work now competes for attention with a dozen other priorities inside those bigger teams.
Who’s Leaving and Who’s Staying
The team’s lead, Dylan Scandinaro, hasn’t left OpenAI — he’s shifted focus to studying the impact of recursively self-improving AI, which is essentially the question of what happens when AI systems start getting significantly better at improving themselves without much human input. Other senior staff from Preparedness have been redistributed into narrower lanes, some focusing specifically on cybersecurity risks and others on biosecurity risks, rather than covering the whole spectrum of dangerous capabilities under one roof.
But the reshuffle doesn’t stop there. Several other well-known names connected to AI safety and long-term thinking at OpenAI have also stepped away recently: ethics lead Chloe Bakalar, Chief Futurist Josh Achiam, and safety chief Johannes Heidecke are all reported to have exited their roles. When you put the team dissolution and these departures side by side, it starts to look less like routine reorganization and more like a broader shift in how much weight OpenAI is putting behind dedicated, long-horizon safety roles.
The Bigger Picture: OpenAI’s Restructuring Wave
The Preparedness team isn’t the only casualty here. OpenAI has also dissolved its AGI readiness team and its superalignment team — two groups that, on paper, existed specifically to think about the long-term, high-stakes future of increasingly capable AI systems. That’s three dedicated safety-adjacent teams either gone or absorbed elsewhere in a fairly short window.
What makes the timing interesting is that all of this is happening while OpenAI is reportedly gearing up for a major IPO. Companies preparing to go public tend to streamline operations, cut duplicate functions, and focus resources on things that drive growth and investor confidence. It’s not hard to see how research teams focused on hypothetical worst-case scenarios might get deprioritized in that kind of environment, even if nobody frames it that way publicly.
There have also been leadership shifts on the business side around the same time: Chief Revenue Officer Denise Dresser announced she’s leaving to pursue other opportunities, with Dali Rajic stepping in as her replacement, and COO Brad Lightcap has revealed plans to exit and start his own venture. None of these moves are necessarily connected to the safety team changes, but together they paint a picture of a company going through a lot of internal change all at once.
What This Means for AI Safety Going Forward
So does this mean OpenAI has stopped caring about dangerous AI risks? Not exactly — the individual specialists are still there, still working on cybersecurity and biosecurity concerns, just inside different organizational boxes. But there’s a real difference between having one team whose entire mandate is catastrophic risk assessment, and having that work spread across teams that also have other things to worry about.
Critics of this kind of restructuring usually point out that dedicated teams tend to have more independence and are harder to quietly deprioritize than tasks handed to already-busy groups. Supporters of the change might argue that embedding risk specialists directly into the teams building the models actually makes the safety work more practical and less siloed. Both arguments have some truth to them — the real test will be whether OpenAI’s future models still get the same level of scrutiny before release, or whether this reshuffle quietly lowers the bar.
For everyday users, this story is a reminder that the AI tools we use every day are built inside companies making constant trade-offs between speed, safety, and business pressure. It’s worth keeping an eye on how these internal shifts play out, especially as AI models keep getting more capable and the stakes of getting safety wrong keep climbing.
| Specification | Details |
|---|---|
| Team Disbanded | OpenAI Preparedness Team |
| Disbanded When | End of July 2026 |
| Reported By | Financial Times |
| Former Team Lead | Dylan Scandinaro (now on recursive self-improving AI impacts) |
| Original Mandate | Assess dangerous AI risks (cybersecurity, biosecurity) and mitigation strategies |
| Related Teams Also Dissolved | AGI Readiness team, Superalignment team |
| Notable Departures | Chloe Bakalar (Ethics Lead), Josh Achiam (Chief Futurist), Johannes Heidecke (Safety Chief) |
| Company Context | Broader restructuring while preparing for a major IPO |
| Category | Details |
|---|---|
| Original Price | Not Applicable — this is a corporate restructuring story, not a product launch |
| Current Price | Not Applicable |
| Bank Card Offer | Not Applicable |
| EMI | Not Applicable |
| Exchange | Not Applicable |
Frequently Asked Questions
The OpenAI Preparedness team was an internal group that studied whether powerful AI models could cause serious, real-world harm — things like cybersecurity threats or biosecurity risks — and worked out ways to reduce those dangers before models were released.
According to Financial Times reporting, OpenAI folded the team at the end of July 2026 as part of a larger reshuffle of its safety-focused groups, moving specialists into other existing teams rather than keeping Preparedness as a standalone unit.
Dylan Scandinaro led the Preparedness team. He has since moved to a role focused on studying the impact of recursively self-improving AI, while other senior members shifted into cybersecurity and biosecurity-specific tracks.
Alongside the Preparedness team shake-up, OpenAI has dissolved its AGI readiness and superalignment teams, and seen several high-profile exits including ethics lead Chloe Bakalar, Chief Futurist Josh Achiam, and safety chief Johannes Heidecke, even as the company gears up for a major IPO.
OpenAI says the safety work continues inside other teams rather than disappearing, but critics worry that folding a dedicated risk-assessment group into broader teams could mean less focused attention on catastrophic AI risks during a period of rapid growth.
Verdict: Should You Be Worried?
Bottom line — the disbanding of the OpenAI Preparedness team isn’t proof that the company has abandoned AI safety, but it’s a clear signal that the org chart around dangerous-capability research is shifting, right as OpenAI juggles an IPO and a wave of leadership exits. Whether this turns out to be smart consolidation or a step back will depend entirely on how future models get evaluated before they ship. Either way, it’s a story worth watching closely.
Want more clear, no-jargon breakdowns of what’s happening in AI and tech? Head over to JatinTechTalks for more updates like this one.
No comments:
Post a Comment