OpenAI disbanded its Preparedness team at the end of July 2026, according to revelations by the Financial Times and WIRED , sparking a new wave of concerns about the future of security in the development of advanced artificial intelligence systems. The team was tasked with assessing whether AI models posed serious risks — such as the potential for them to “get out of control” and hack into other companies — and developing ways to mitigate those risks . The move is part of a broader reorganization of the company ahead of its expected IPO .
According to reports, the Preparedness were not eliminated entirely, but were distributed to existing sector teams: responsibilities for biothreats and security were transferred to specialized teams that already operated within the company. The question critics are asking is not whether security work disappeared, but whether the absorption of an independent disaster risk assessment team into product groups weakens central oversight.
See also: GPT-5.6-Cyber: OpenAI lowers the bar on vulnerability research
Preparedness team leader Dylan Scandinaro , who was hired by competitor Anthropic in February 2026, remains at the company but is now focusing on the implications of recursive self-improving AI — one of the most critical and controversial areas of research in the space.
OpenAI's Preparedness Framework: History and Importance
OpenAI created the Preparedness Framework in 2023 and updated it in 2025 , with the goal of identifying, assessing, and managing serious risks from advanced AI systems . The framework covers high-impact areas such as cybersecurity , biological threats , persuasion , and autonomous or self-improving model behavior. It was designed to create internal “gateways” for releases that may exceed risk thresholds. OpenAI’s public governance documents from May 2026 still describe the Preparedness Framework as active and as the basis for assessing serious risks, responding to incidents, and bringing in external expertise.
But the reality looks different. In recent years, the company has gradually abandoned its goal-driven model, disbanding teams working on AGI (Artificial General Intelligence) and superalignment. Also recently, chief ethics officer Chloé Bakalar, chief futurist Josh Achiam , and chief security officer Johannes Heidecke. These departures have fueled concerns from critics like Jan Leike, who resigned from OpenAI in 2024 and told the Financial Times that the company is ignoring security in favor of creating “shiny products.”
See also: OpenAI Astra: Pause on most powerful AI model due to critical cyber capabilities
Competitive context and the pressure for Preparedness
OpenAI operates in a highly competitive frontier AI, where Google and Anthropic remain its main rivals, while Chinese and other competitors compete on cost, framework length, and specialized use cases.
Competition is intensifying around product bundling and ecosystem control. Google ’s Gemini has benefited from integration with Android , while Anthropic ’s Claude is gaining market share in some markets, particularly among users looking for powerful reasoning and enterprise workflows. This competitive pressure partly explains why OpenAI may be under pressure to move quickly in product delivery, even as it tries to maintain a credible security posture.
The broader industry trend is toward integrating security assessment into product teams rather than keeping it isolated to a single, central team. OpenAI’s reorganization seems consistent with that pattern, but critics worry that it could also dilute accountability, because no one team now has primary responsibility for stopping dangerous releases.
Cybersecurity Risks and the Future of Preparedness
One of the most immediate risks the Preparedness team faced was autonomous cyber abuse. AI models can potentially accelerate attacks phishing , vulnerability discovery , or malware customization — this is one of the clearest risks in frontier AI development . Transferring this responsibility to a cybersecurity team that is simultaneously under pressure to deliver products raises legitimate questions about maintaining the independence of the assessment.
See also: Qwen3.8-Max: Alibaba challenges OpenAI and Anthropic
For businesses using OpenAI tools , the change has practical implications. Cyber and bio risks can impact business trust, compliance, and procurement processes. If security assessment becomes more distributed, customers may demand clearer evidence that high-risk assessments are still being performed before new models are deployed . This is particularly important for organizations operating in regulated sectors such as healthcare, finance, and critical infrastructure.
