Proofpoint’s agentic workspace pivot puts AI agent security in scope

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Proofpoint's blog post: The modern workspace now includes AI agents alongside people, and security has to govern both in the same collaboration and data environment.

Why it matters: IAM, PAM, data security and SaaS governance teams need to treat AI assistants and agents as part of the workspace threat model, not as an edge case outside existing controls.

Context

The core governance gap is that workspace security was built for human users, while collaboration systems now include AI assistants and agents that can act, retrieve, generate and expose information at machine speed. That changes the control problem from protecting accounts alone to governing interaction, delegation and data movement across both people and non-human actors.

In this model, the security boundary is not just login and access. It is the combination of identity, content, collaboration and data handling, which means NHI governance starts to overlap with human IAM, DLP, DSPM and SaaS control design.

Proofpoint is describing a direction that is becoming typical rather than exceptional: enterprise work is moving toward shared human and agent workflows, and the organisations that still separate those controls will carry the most governance risk.

Key questions

Q: How should security teams govern AI agents in shared workspaces?

A: Security teams should treat the workspace audience as part of the authorization decision. If the agent can be seen by people with different entitlements, the system must check whether every recipient is allowed to receive the data before the response is generated. Otherwise, the agent can disclose information correctly retrieved but incorrectly exposed.

Q: Why do AI agents increase workspace security risk for data teams?

A: AI agents can retrieve, transform and redistribute sensitive content at machine speed, which increases the chance that a single policy failure becomes many exposures. They also blur the line between authorised use and unintended propagation, so data teams must control not only the file or message, but the agent's ability to act on it.

Q: What are the warning signs that agentic workspace controls are failing?

A: The strongest warning signs are unmanaged assistants, unclear data boundaries, excessive agent permissions and weak logging for agent-driven actions. If security teams cannot say which agents touched which data or why a workflow was allowed, the governance model is already operating outside its intended boundary.

Q: What should organisations do when human and non-human workflows share the same workspace?

A: Organisations should create a joint governance model for identity, collaboration and data security. That model should define approval, review, monitoring and offboarding for both people and agents, because shared workspaces fail when controls assume only one type of actor can move information or make decisions.

Technical breakdown

Why the agentic workspace changes the control surface

The agentic workspace is a collaboration environment where humans, copilots and AI agents all touch the same messages, files and SaaS workflows. The technical change is not just more automation, but a new class of runtime actor that can read, summarise, generate and forward information at speed. That breaks assumptions in classic workspace security, which usually treats the user as the only active decision-maker and the application as a passive container. Once an agent can retrieve context, trigger actions or surface content across tools, the control surface expands from identity verification to permissioned interaction, content governance and downstream data handling.

Practical implication: security teams need to inventory which AI agents can access collaboration and data tools, not just which human accounts can.

How prompt injection and social engineering converge in shared workspaces

Prompt injection is the AI-era analogue of social engineering, but the target is the agent's instruction chain rather than a human's judgement. In a shared workspace, the same message, attachment or document can influence both a person and an agent, which creates coupled risk paths. A human may be tricked into approving or forwarding something, while an agent may be manipulated into retrieving data, summarising sensitive content or taking an unsafe action. That makes workspace governance dependent on both content inspection and policy enforcement at the point where agents interpret input, not only where users authenticate.

Practical implication: policy has to inspect agent inputs and outputs as governed events, not just human communications.

Why data loss controls must extend to AI-generated and AI-consumed content

Data loss in an agentic workspace is no longer limited to an employee attaching the wrong file or sending the wrong message. Agents can reproduce, transform and redistribute sensitive information at high volume, which means the data path becomes more dynamic and more difficult to reason about after the fact. DLP and DSPM remain relevant, but they now need to understand which identities, including non-human ones, are allowed to see, transform and propagate specific data classes. Without that, controls may observe leakage only after the agent has already amplified it across email, chat and SaaS workflows.

Practical implication: data controls should classify agent access and agent-generated content with the same rigour used for regulated human workflows.

Breaches seen in the wild

• Microsoft verified publisher OAuth phishing 2022: Malicious OAuth apps with a fraudulently obtained Microsoft verified publisher badge tricked UK users into granting mailbox access in 2022.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

The agentic workspace creates a governance overlap that legacy IAM never had to solve. Traditional identity programmes were built around humans requesting access and systems enforcing it at sign-in or session creation. Once AI agents operate inside the same workspace, identity decisions move into runtime behaviour, delegation and content handling. That makes the issue more than access management. It becomes a control problem across human identity, non-human identity and data movement, with direct implications for IAM, PAM, DLP and DSPM teams. Practitioners should treat this as a governance redesign, not a point-product adjustment.

Human-centric security remains necessary, but it is no longer sufficient. Proofpoint's framing reflects a broader market truth: collaboration security is becoming the control plane for both people and agents. The security objective is shifting from protecting users from messages to protecting workflows from manipulation. That means the old separation between user risk and system risk is breaking down. Teams that keep agent controls outside their workspace governance model will miss the point where misuse becomes operational.

Agent visibility is the new prerequisite for enforceable policy. If security teams cannot see which assistants or agents are present, what data they touch and which actions they can trigger, policy will be largely symbolic. The named concept here is workspace identity convergence: the point at which human identity, NHI governance and collaboration control have to be managed together. Practitioners should design for this convergence now, because retrofitting it after broad agent adoption will be far more expensive.

Data security becomes the practical enforcement layer for agent governance. AI agents do not just need authorisation, they need tightly bounded exposure to data classes and workflow steps. That makes DLP and DSPM central to the agentic workspace discussion, not peripheral. Organisations that already have strong data classification and access segmentation will adapt faster than those relying only on user-centric approval chains. The implication for identity teams is clear: agent governance fails where data governance is weak.

The market is moving toward integrated human and agent security platforms. The vendor ecosystem is converging on a single operational reality: collaboration, identity and data controls are being pulled together by AI use cases. That does not eliminate the need for specialised controls, but it does change buying criteria. Practitioners should evaluate whether their current stack can enforce policy across both people and agents, or whether it still assumes a human-only workspace.

What this signals

workspace identity convergence: human identity, non-human identity and collaboration governance are converging into one operational problem. Teams that still separate agent controls from workforce controls will struggle to enforce policy consistently across email, chat and SaaS workflows.

The next governance step is to decide which agent actions are acceptable before data ever reaches the workspace layer. That pushes control design upstream into authorisation, logging and data segmentation, where human IAM and NHI governance now intersect most visibly.

For practitioners

• Map AI assistants and agents to workspace permissions Identify every assistant, copilot and autonomous agent that can read, summarise or act inside email, collaboration and SaaS systems. Tie each one to the specific permissions it uses, the data classes it can see and the actions it can trigger.
• Extend DLP and DSPM to non-human workflows Review whether current content inspection and sensitive-data controls can distinguish between human sharing and agent-driven transformation, forwarding or retrieval. Prioritise workflows where an agent can amplify a mistake across multiple systems.
• Harden prompt and content handling policies Treat prompts, replies, attachments and summaries as governed inputs and outputs. Apply controls that reduce the chance of prompt injection, unsafe retrieval and accidental disclosure when an agent processes user content.
• Build a shared human-and-agent governance model Assign ownership for agent onboarding, access review, logging and offboarding across IAM, collaboration security and data security teams. Document who approves agent capabilities and who can revoke them when behaviour changes.
• Test incident response for agent misuse scenarios Run exercises that assume an agent has exposed sensitive content, forwarded the wrong data or executed an unsafe action inside a workspace workflow. Validate containment steps, logging coverage and escalation paths across identity and data owners.

Key takeaways

• The article frames AI assistants and agents as part of the workspace threat model, not as peripheral automation.
• The main security challenge is controlling how data, prompts and actions flow across both human and non-human actors.
• Teams need governance that joins IAM, collaboration security, DLP and DSPM instead of treating them as separate problems.

Key terms

• Agentic Workspace: An agentic workspace is an environment where humans, AI agents, and connected tools operate in shared workflows. The security challenge is that actions may be distributed across multiple identities and systems, making scope, traceability, and accountability harder to maintain.
• Workspace Identity Convergence: Workspace identity convergence is the point where human identity, non-human identity and collaboration governance must be managed as one control problem. It emerges when agents and users share permissions, data paths and workflow authority across the same SaaS environment.
• Prompt Injection (Agentic): An attack where malicious instructions are embedded in content that an AI agent reads, causing the agent to execute unintended actions using its own legitimate credentials. A primary vector for agent goal hijacking and identity abuse.
• Data Security Posture Management: Data Security Posture Management, or DSPM, is the continuous discovery and monitoring of where sensitive data lives, how it is exposed, and where policy gaps exist. Its value rises when it feeds remediation rather than generating findings alone, especially in environments where AI expands the number of data paths.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, agentic AI identity and machine identity security. It is designed for practitioners who need to translate identity concepts into operating controls across real programmes.