December 2025 Summaries

Generative AI's integration into modern enterprises is significantly supported by platforms like Google Vertex AI, which facilitates the large-scale building, tuning, and deployment of AI models. Vertex AI provides a unified environment for organizations to experiment and operationalize AI, supporting various models and third-party ecosystems. However, this capability introduces complex identity governance challenges, as model inputs may contain sensitive data and permissions related to model execution and tuning can lead to security risks. The platform's integration with Google Cloud Resource Manager allows cross-project and multi-region deployments, which can create new access propagation paths if not properly governed. Effective identity governance is crucial, involving replacing standing access with just-in-time invocation, enforcing separation of duties, establishing clear identity provenance, and governing cross-project and cross-region usage. By addressing these governance challenges, organizations can harness the power of Vertex AI while ensuring secure and compliant AI adoption, balancing rapid innovation with robust security measures.

The P0 Authz Control Plane for Agents is designed to help developer and security teams manage access control for agentic applications that interact with internal data sources via a chat interface. By providing API endpoints, it enables the restriction of access to various MCP tools based on user roles and facilitates data access policy definition at multiple levels within data sources like Postgres. P0 supports Just-in-Time Access by allowing approval requests for temporary access, ensuring that only role-appropriate tools are accessible to users. This approach addresses the challenges posed by agentic AI apps, which have more freedom in executing data actions compared to traditional web applications. P0 implements a two-layer control system: the MCP Tool Access Control filters tools based on user roles, while the Data Layer Access Control ensures users can only execute authorized SQL queries. P0's setup includes defining policies in its Policy Studio, which map roles to tools and evaluate SQL query permissions, thereby narrowing user access to only the data they are permitted to see. Additionally, P0 supports a Just-in-Time and Human-in-the-Loop system, allowing AI to autonomously request access to roles or tools, subject to human approval, thereby enhancing security without compromising functionality.

Google Vertex AI has become a central platform for enterprises to build, tune, and deploy Large Language Models and multimodal AI workloads at scale. It supports a variety of models, including Gemini, PaLM, and Llama, while providing a unified environment for AI experimentation and operationalization. However, as organizations expand AI model usage, Vertex introduces new identity governance challenges, as access to AI models can expose sensitive data and lead to compliance issues. Key risks include runtime invocation, model lifecycle management, and cross-project access, which require rigorous identity management and governance controls. To mitigate these risks, organizations should enforce just-in-time access, separation of duties, and clear identity provenance. Moreover, cross-region and cross-project model sharing necessitate strong governance to prevent unauthorized access and data residency violations. Effective identity governance ensures that as organizations scale AI on Vertex, they can do so securely, aligning with corporate governance policies and minimizing potential risks.

The P0 AuthZ Control Plane for Agents is designed to enable developer and security teams to manage access for agentic applications that connect to internal data sources, such as databases, via chat interfaces. It provides API endpoints that control access to MCP tools and can restrict access based on user roles, with the ability to define data access policies at granular levels within the data source. The system also supports Just-in-Time Access, enabling temporary and scoped permissions through human approval processes. In agentic apps, AI can execute actions on data sources, and P0 ensures that only role-appropriate tools are accessible, eliminating risks associated with unauthorized tool selection. By integrating a robust access control framework that aligns with the RBAC model and deploying defense-in-depth strategies, P0 narrows user access to permissible data and actions, while also allowing for human-in-the-loop workflows to be initiated for specific access requests.

An ISMG interview with Neha and Kelsey highlights the challenges and evolution of privileged access management (PAM) in a rapidly changing digital landscape. The discussion emphasizes the shift from authentication to authorization, identifying the need for enterprises to decide what actions users or systems can perform at any given moment, rather than just verifying identities. This shift is crucial as traditional tools struggle to cope with the dynamic nature of modern IT environments, such as the proliferation of cloud services and non-human identities. The interview also underscores the problem of tool fragmentation, where disparate systems manage different aspects of access control without providing a cohesive view, leading to increased operational complexity. The recent announcement about ZSP addresses these issues by advocating for a unified model of identity and access, aiming to automate decision-making and minimize static credentials. The conversation reflects broader industry trends and customer demands for governance automation and streamlined access management to keep pace with accelerating agentic systems.

In an interview with ISMG, Neha and Kelsey highlighted the evolving challenges of privileged access management (PAM) in a world where cloud services, dynamic workloads, and non-human identities are increasingly common. They emphasized the shift from authentication to authorization, noting that the main issue is not verifying identities but determining what actions these identities should be authorized to perform. Legacy tools, designed for static environments, are inadequate for the modern landscape, and the fragmented nature of existing solutions complicates the management of access and entitlements. The interview aligns with insights from CISOs and analysts, stressing the need for unified models of identity and access, where governance automation plays a crucial role in managing diverse identity types and streamlining the access model. This approach, encapsulated by the recent ZSP announcement, aims to simplify access management and eliminate standing privileges by integrating discovery, risk assessment, policy enforcement, and lifecycle management into a cohesive system.

The final installment of a three-part series on modern privileged access management (PAM) explores how its evolution enhances security and productivity by protecting more systems and identities. The text emphasizes the growing threat of identity-centric attacks and the complexities introduced by cloud technologies, which can create blind spots and inconsistencies in privileged access control. It discusses the need for a more integrated PAM approach, focusing on zero standing privileges and just-in-time access, while also highlighting the importance of automation, governance, and an API-first strategy for scalability and security. The article stresses the necessity of understanding the current PAM landscape before transitioning to a future architecture, recommending a strategic migration plan based on pain point assessments and prioritizing high-risk systems. The author, Simon Moffatt, brings extensive expertise in identity and access management, providing insights into effectively managing privileged access in modern environments.

The text explores the evolution of Privileged Access Management (PAM) in the context of modern security challenges and cloud adoption, emphasizing the need for an API-led access management approach to enhance both security and productivity. It highlights the increase in identity-centric attacks and the complexities introduced by reliance on cloud technologies, which necessitate more robust privileged access controls. The article provides recommendations for adopting and migrating to a more integrated PAM approach, focusing on governance, automation, and control through zero standing privileges and just-in-time access. It stresses the importance of understanding both the current PAM landscape and designing a future architecture that addresses high-risk areas, while also planning migrations with a focus on prioritizing business requirements, risks, and rewards. The author, Simon Moffat, brings his extensive experience in identity and cyber security to guide organizations in strategically managing privileged access and mitigating security risks.

AI agents are becoming integral in modern developer teams for building, automating, and scaling processes, with platforms like AWS Bedrock and Google Vertex facilitating their direct interaction with cloud resources and applications. However, their autonomy introduces risks related to static credentials and broad access permissions, which are often inadequately monitored and audited. P0's Production Access Control Plane addresses these issues by extending its least-privilege, just-in-time access model to AI agents, allowing organizations to secure, govern, and audit agentic access. This is achieved through secondary authorization controls based on the human identity interacting with the agent, enabling a just-in-time approval process for access requests and rigorous session-level auditing. This capability is part of the P0 platform, offering a centralized control plane for managing all identities, ensuring security and consistency across major cloud providers, and supporting security, identity, and platform teams in deploying AI applications responsibly. The new features are available in early access for select partners, inviting collaboration to enhance access governance in AI deployments.

P0 introduces a new feature set that enhances the security and governance of AI agents by implementing a just-in-time (JIT) access model, extending beyond human users to first-party AI agents. These agents, which interact autonomously with cloud resources and applications, traditionally operate with static credentials, posing security risks due to broad permissions and lack of monitoring. P0 addresses these issues by allowing organizations to apply secondary authorization controls based on the human identity interacting with the agent, narrowing permissions to specific scopes required by the end-user. This approach introduces a human-in-the-loop approval workflow for privileged access, ensuring that every session is recorded and auditable, thus maintaining accountability and security. The P0 platform offers a centralized control plane for all identities, enabling consistent policy enforcement across major cloud providers and services, and it aims to balance productivity with security by preventing identity sprawl while allowing seamless deployment of AI-driven productivity applications. This capability is currently available in early access for select partners interested in experimenting with first-party agentic applications and exploring secure AI scaling in production environments.