OpenAI Forms Safety and Security Committee With Aleksander Madry as Head of Preparedness
OpenAI’s new Board Safety and Security Committee formalizes oversight of frontier-model risks, with Aleksander Madry leading the company’s Preparedness work.
OpenAI formally created a Board Safety and Security Committee in May 2024, placing safety and security oversight more directly within its board structure as the company develops increasingly capable AI systems. The committee’s technical and policy contributors include Aleksander Madry, OpenAI’s Head of Preparedness, whose team is intended to connect frontier-model capability assessment, evaluations, and red-teaming.
The move is significant because it turns a broad commitment to AI safety into a defined governance and review process. In its official announcement of the Board Safety and Security Committee, OpenAI said the group would initially evaluate and further develop the company’s safety processes over 90 days, followed by public updates after the Board’s review.
Madry’s role sits at the technical end of that effort. OpenAI described the dedicated Preparedness team as work spanning risks from near-term systems through potential artificial general intelligence, or AGI-level, capabilities. The public discussion around his appointment also referenced the much longer-term safety and security challenges associated with superintelligence. The verified organizational change, however, is the formation of the committee and the Preparedness function, not the release of a new model, safety standard, or developer product.
How OpenAI’s safety structure is organized
The Safety and Security Committee was chaired by OpenAI chief executive Sam Altman. Its board participants included Bret Taylor, Adam D’Angelo, and Nicole Seligman, while OpenAI named a group of internal technical and policy experts to support the work. In June 2024, OpenAI updated its announcement to add retired U.S. Army General Paul M. Nakasone to the committee.
The committee and Preparedness team have related but distinct purposes. One provides a board-level route for reviewing safety and security processes. The other focuses on assessing model capabilities and risks through technical evaluation and red-teaming.
| Structure | Primary function described by OpenAI | Relevant leadership |
|---|---|---|
| Board Safety and Security Committee | Evaluate and develop OpenAI’s safety processes, with Board review and public updates | Chaired by Sam Altman; includes board members and internal experts |
| Preparedness team | Connect capability assessment, evaluations, and red-teaming for frontier models | Aleksander Madry, Head of Preparedness |
OpenAI also identified Lilian Weng, Head of Safety Systems; John Schulman, Head of Alignment Science; Matt Knight, Head of Security; and Chief Scientist Jakub Pachocki among the experts serving on the committee. Together, those roles indicate that the committee was designed to draw on security, model-safety, alignment, scientific, and preparedness expertise rather than treat frontier risk as a single technical problem.
What the announcement changes, and what remains open
The confirmed change is organizational: OpenAI established a formal committee, set an initial review period, and identified the people and functions involved. That matters because technical risk assessment can be elevated into a recurring board-level process rather than handled only within individual research or product teams.
At the same time, the announcement did not publish a new external safety benchmark, a new API control, or a specific developer tool. It also did not detail every standard the committee would adopt. Readers should therefore view this as a governance and risk-management initiative, not as a direct change to how developers access or use OpenAI products.
For businesses using AI, the practical signal is more indirect. As frontier-model providers formalize evaluation, red-teaming, security, and oversight processes, teams adopting AI should similarly distinguish between promising use cases and the operational risks around them. A customer-facing workflow, for example, may need testing for inaccurate outputs, while an internal automation that handles sensitive business information may require careful choices about access and review.
Useful questions for decision-makers include:
- What tasks can AI assist with safely under human review?
- What information should not be placed into an AI workflow without appropriate controls?
- How will the team test outputs before using them in customer, financial, or operational processes?
- Who is accountable for monitoring an AI-enabled process after it is deployed?
These are not new requirements created by OpenAI’s committee. They are practical adoption questions that become more important as AI tools become more capable and more integrated into day-to-day work.
For companies moving from ad hoc experimentation to repeatable AI use, Scalevise can help turn promising ideas into practical, tested workflows. Our AI implementation and use-case consulting helps teams prioritize high-value opportunities, assess operational fit, and plan integrations without adding unnecessary complexity. That gives decision-makers a clearer path from isolated AI tools to processes that save time and support reliable work. Request an AI consultancy discussion with Scalevise.
Frequently Asked Questions
What is OpenAI’s Safety and Security Committee?
It is a board-level committee OpenAI formed in May 2024 to evaluate and develop the company’s safety processes. Sam Altman chaired the committee, which included board members and internal technical and policy experts.
What is Aleksander Madry’s role at OpenAI?
Aleksander Madry was named Head of Preparedness. OpenAI said the Preparedness team connects capability assessment, evaluations, and red-teaming for frontier models.
Did OpenAI announce a new safety product or developer tool?
No. The announcement described a governance and safety-process initiative. It did not introduce a new external safety benchmark, API control, or developer tool.
Why does the committee matter for businesses using AI?
It highlights the importance of evaluating risks alongside AI capabilities. Businesses can apply the same principle by testing AI outputs, limiting sensitive-data exposure, and keeping appropriate human review in important workflows.
Conclusion
OpenAI’s Safety and Security Committee formalizes board-level oversight of frontier AI risks, while Aleksander Madry’s Preparedness team provides a technical link between model assessments, evaluations, and red-teaming. The announcement does not change a specific product or API, but it makes OpenAI’s safety and security effort more structured as model capabilities advance.