According to The Verge, citing people familiar with the matter,OpenAI has disbanded its Preparedness Team. The dissolution of this team, which was responsible for assessing the potential dangers of cutting-edge AI models in areas such as cybersecurity, chemical/biological risks, and autonomous capabilities, means that the last independent security assessment barrier within OpenAI is disappearing.
What does a security preparedness team do?
The Preparedness Team was established in 2023 and is another independent security department established by OpenAI outside the "Super Alignment" team. Its core task is not theoretical research;Risk assessment at the practical operational level: Before the model is released, use red team attacks to test whether the model can be used to create dangers, whether it can independently break through the sandbox restrictions, and how it performs in network security attack scenarios.
In June 2026, it was this team that discovered that the GPT-5.6 series models could not rule out "critical" level network risks during testing, which directly caused OpenAI to postpone the public release of some models. The existence of this team is the most concrete evidence of OpenAI’s external commitment to “security first”.
Why was it disbanded?
At present, OpenAI has not officially responded publicly, but based on information from various parties, there may be three reasons:
- Commercialization pressure: OpenAI's enterprise revenue has exceeded consumer revenue for the first time. B-side customers need a "stable and available" model rather than a "delayed release due to security considerations" model. The existence of security teams is becoming a bottleneck in the speed of product releases.
- organizational restructuring: OpenAI may disperse security assessment functions among various product teams rather than retaining an independent central review agency. However, the effectiveness of this "decentralized security" model is highly controversial in the industry.
- brain drain: Two senior executives from OpenAI have resigned in the past week. Core members of the security team may also be poached by competitors, rendering the team effectively unsustainable.
Industry reaction: Concerns and doubts
After the news broke, the response from the AI safety community was swift and fierce. Several former OpenAI security researchers expressed concerns on social media——“When security teams are viewed as an impediment to product launches rather than a safeguard, the entire industry is heading in a dangerous direction.”
It is worth noting that Anthropic CEO Dario Amodei mentioned in an interview at the same time,The crisis of trust in the AI industry is “fundamentally a crisis of trust”. When OpenAI accelerates on the road to commercialization and weakens its internal security mechanisms, users and regulators have reason to question: How much weight are left in those commitments about "responsible AI"?
What does it mean for the average user?
In the short term, users may see a faster pace of model releases and fewer feature limitations. But in the long run,AI models lacking independent safety review are like high-speed trains without brakes——The faster you run, the greater the risk. This is especially noteworthy for enterprise users who are deeply embedding AI into business processes.
Are you concerned about AI industry trends and tool reviews? access AI Dash, helping you make smarter AI decisions.
Pingback: AI Is Driving Up Medical Bills: Hospitals Spent $942 Million More on Claims Submissions Using AI Over Two Years — AI Dash