As organizations increasingly integrate AI tools, employees are utilizing public platforms with sensitive data, yet many Chief Information Security Officers (CISOs) are navigating this landscape without a solid framework. The Clusif has released a landmark guide from its working group on “Secure AI Usage in Organizations,” outlining initial security measures prioritized into three operational postures. This guide is intentionally designed for immediate activation without requiring significant organizational transformation and can be integrated into existing governance and risk management frameworks.
Founded in late 2025, the working group behind this initiative comprises CISOs, legal experts, consultants, and risk managers from major corporations, including Sanofi and Suez. This diverse composition reflects a methodological choice: securing AI usage encompasses legal, technical, risk management, and supplier management aspects. None of these elements can be effectively addressed in isolation. The document speaks directly to CISOs and cybersecurity leaders facing varied situations and does not claim to constitute compliance with the AI Act or other regulations.
The prioritization methodology rests on four interconnected criteria. Firstly, the organization’s AI maturity level is assessed based on governance adequacy and inventory of deployed systems. Secondly, operational pressure is determined by business demands and the proliferation of uncontrolled AI usage. Thirdly, urgency and risk exposure are evaluated, notably in the face of actual incidents or imminent regulatory scrutiny. Lastly, both immediate impact and feasibility are crucial, with only those measures yielding visible effects without extensive redesign being considered. This selection framework leads to three distinct postures, each linked to a unique sequence of prioritized actions.
Establish Foundations Before Implementing Controls
For a CISO tackling a loosely defined AI perimeter, the Clusif emphasizes governance over technology. The 11 priority measures within the first posture commence with forming an AI governance committee comprising the CISO, DPO, legal representatives, procurement, and key business leaders. This is followed by the formalization and immediate dissemination of an AI usage charter. Mapping AI systems, which includes tools, processed data, deployed agents, and identified owners, is addressed only in fourth place after raising employee awareness.
This ordering reflects the working group’s conviction that deploying technical controls within an organization without a shared visibility on its own usage is akin to securing an unknown perimeter. Furthermore, the document highlights that implementing these measures requires collaboration beyond the CISO alone; it necessitates support from legal, DPO, and IT departments to establish a culture of shared responsibility.
Shadow AI: Inventory, Cataloging, and Mitigating Risks
The second posture targets CISOs whose organizations are already using AI tools extensively without centralized oversight. The primary risk identified is a loss of control over actual usage, directly exposing organizations to data breaches and regulatory non-compliance. The Clusif suggests three successive approaches: gaining visibility through a comprehensive inventory of AI tools used; channeling by referencing and communicating approved solutions; and controlling by monitoring real usage and blocking unauthorized access.
Among the ten measures in this posture, reviewing the terms of service and contractual clauses with AI suppliers ranks fifth. The document underscores frequently overlooked contractual parameters during the adoption of consumer tools or third-party solutions, such as support procedures in the event of an incident, response availability, language of communication, and effective service level agreements (SLAs). These factors directly influence the organization’s ability to respond to incidents involving external AI tools.
In Crisis Situations, Contain Before Structuring
The third posture, dubbed the “CISO Firefighter,” addresses ongoing operational emergencies. The eight measures follow a crisis management logic tailored to the specifics of AI. The first step is to contain the situation by activating controls on the input and output of the affected AI systems. Next, a rapid evaluation of high-risk usages is conducted. Following this, access to implicated systems is cut off, investigative elements are collected, and crisis communication with stakeholders is organized.
Two measures distinguish this posture from traditional incident response protocols. The implementation of AI security testing, including prompt injection and jailbreak exercises, takes place during the technical hardening phase following containment. Finally, the crisis review, which concludes the sequence, marks the transition to a structured governance posture. It is from the lessons learned during the incident that the CISO can begin the shift to the first posture, basing decisions on verified facts rather than theoretical estimates.
The Clusif notes that the proposed measures are deliberately positioned at a macro level, given the rapid evolution of involved technologies. The operational implementation of these strategies will vary based on the technical environment available in each organization. A second deliverable from the working group is anticipated, focusing on the evaluation of AI solutions and implementations, employee awareness, and enhanced governance of AI within organizations.



