This content is currently locked.

Your current Infotech subscription does not include access to this content. Contact your account representative to gain access to Premium SoftwareReviews.

Contact Your Representative
Or Call Us:
+1-888-670-8889 (US/CAN) or
+1-703-340-1171 (International)

Forcepoint AI Security: AI Security Built on a Data Security Foundation

Technology Note By: Jon Nelson, Infotech

Executive Summary

Forcepoint is extending its converged data security platform (DSPM, DDR, DLP) into a dedicated AI security offering, launching in general availability in September 2026. The new capabilities – AI detection and response, shadow AI discovery and control, and an AI agent gateway – sit on top of the existing Data Security Everywhere foundation rather than as a bolt-on product. Whereas other AI security entrants build agent governance and adversarial testing from scratch, Forcepoint extends a classification and enforcement engine it has spent two decades tuning. It also brokers all agent-to-application access rather than relying on agents to self-report inherited permissions.

Forcepoint AI security is a strong fit for organizations already running, or considering, its data security platform, and for buyers whose AI risk extends beyond sanctioned SaaS apps into email, endpoint, and network channels. It is a weaker fit for buyers seeking a single, fully native platform including model-level adversarial testing or for those evaluating pure-play AI security specialists with more mature GRC and AIBOM capabilities.

Introduction

No CISO is choosing how AI enters their organization. Leadership mandates the transformation, and the deployment happens everywhere at once. Engineering stands up agents and MCP servers. A business unit buys a copilot. Someone connects Claude to a production data set. Some of it is sanctioned. Most of it is not. By the time the security team assembles a picture of what is running, that picture is already out of date.

I hear a version of this every week from members. In one recent case, an infrastructure leader asked to build an AI security framework pulled an initial inventory of 66 AI applications in his environment. Within a month, better tooling put the number at 187, and he still assumed the list was incomplete. His team had already sunk 30 to 40 hours into the manual survey and still could not be confident they had found everything, because the footprint kept growing underneath them. The same organization bought a single copilot agent and found nineteen running weeks later, several consuming token budget with no oversight. Shadow AI is the default state, not an edge case, and the security team is told to go secure it after the fact.

What this produces is a buyer under three simultaneous pressures: AI is already everywhere, accountability for securing it still rests with the CISO, and the team's understanding of how these systems work rarely matches what they are being asked to govern. No platform fully resolves that. The best a platform can do is give a stretched security team the visibility and enforcement needed to become effective on an AI footprint that is already sprawling around them.

Forcepoint is entering this space from the data security corner, extending a converged DSPM, DDR, and DLP platform into AI-specific controls. What that combination delivers, where it differentiates, and which buyers should evaluate it is the subject of this note.

Product Overview

Forcepoint AI security is delivered as an extension of the Forcepoint Data Security Cloud, the same platform that carries its converged DSPM, DDR, and DLP capabilities. It is not a standalone product. The AI controls draw on the classification engine (AI mesh) and enforcement infrastructure already underneath the data security platform and inherit its deployment flexibility across SaaS, private cloud, on-premises, and air-gapped environments.

The offering is organized around three pillars. AI detection and response provides API-based monitoring across sanctioned AI applications (Claude, ChatGPT Enterprise, Microsoft Copilot, AWS Bedrock), auditing prompts, responses, and file operations with sensitive-data detection and remediation actions. Shadow AI discovery and control uses web-gateway and endpoint telemetry to identify unsanctioned AI use, distinguish corporate from personal accounts, and enforce allow, block, and restrict policies on data moving to AI tools. The AI Agent gateway is an MCP-based enforcement layer that sits inline between agents and the business applications they act on, brokering access so no agent holds application credentials directly.

Pricing is consumption-based; buyers are entitled to the full set of enforcement points and charged on usage, described on the briefing as weighted to data volume rather than per enforcement point.

Features and Capabilities

Core Features

Forcepoint AI security organizes its capabilities into three pillars, all sitting on the existing data security platform and its AI mesh classification engine. The table summarizes coverage. Figures the vendor provided but that were not independently validated are marked vendor-reported, and capabilities described as roadmap or early access are flagged as such.

Capability

Coverage

AI detection and response (sanctioned AI)

API-based monitoring across sanctioned AI applications (Claude, ChatGPT Enterprise, Microsoft Copilot, AWS Bedrock). Audits prompts, responses, file uploads, and downloads, with sensitive-data detection via inheritance from cloud and on-premises DLP policy. Remediation actions include audit, quarantine, safe copy, and delete conversations. Microsoft agentic activity detection in OneDrive and SharePoint.

Shadow AI discovery and control (unsanctioned AI)

Discovers unsanctioned AI apps and websites via web-gateway telemetry, with a shadow AI table capturing app/URL, category, risk score, user, department, and activity count. Differentiates corporate from personal account usage. Enforces allow, block, and restrict policies on uploads and downloads, including blocking personal accounts. Endpoint-agent visibility beyond the web gateway is described as planned, targeted around end of year.

AI agent gateway (agentic AI)

MCP-based enforcement layer sitting inline between AI agents and SaaS MCP servers. No agent holds direct application credentials; all access is brokered. DLP and access-control policies applied to every agent-to-application transaction, with allow or deny down to specific tools and tool actions. Initial MCP server connectors for Salesforce, Atlassian, and Microsoft. Human-in-the-loop approval and immutable audit trail described as built in.

Model integrity and adversarial testing

Delivered through a partnership with F5 rather than natively, covering model red teaming and adversarial resilience. Whether this is included in the platform license or priced separately was not confirmed.

Platform foundation

AI governance dashboard (AI tool landscape, sensitive data exposure, policy effectiveness, MCP gateway, spend governance), guided first-time-use workflow, and ARIA, a natural-language admin assistant backed by a RAG pipeline into product documentation.

Deployment

Inherits the data security platform's SaaS, private cloud, on-premises, and air-gapped options.

Differentiating Features

Brokered agent access, not just agent monitoring. Most AI security products detect what an agent did and log it. The AI agent gateway sits inline and brokers the access itself, so no agent holds direct application credentials. Every agent-to-application call is inspected and authorized against DLP and access-control policy before it reaches the business system, down to specific tools and tool actions. Forcepoint demonstrated this on the briefing: a Claude instance configured with read-only Salesforce access was prompted to write, and the gateway blocked it. That is credential isolation, not after-the-fact visibility, and it addresses the over-privileged-agent problem directly rather than reporting on it.

Cross-channel shadow AI enforcement, not just cloud-side visibility. Many AI security products discover shadow AI through cloud or SaaS API connections. That means they see sanctioned apps but miss data leaving over the web or endpoint. Forcepoint pulls shadow AI telemetry from its web gateway and enforces inline on data heading to unsanctioned tools. It distinguishes corporate from personal accounts, blocking personal-tenant use while allowing the sanctioned one. This comes from its existing DLP placement in the web and endpoint channels, not a new AI-specific sensor, so its value tracks directly to how much AI risk lives outside sanctioned SaaS. The current lean on the web gateway is worth noting: broader endpoint-agent coverage is roadmap, not shipping.

Adversarial testing through a specialist partner, not a native checkbox. Model red teaming and adversarial resilience come through a partnership with F5 rather than a built-in Forcepoint feature. At first glance that looks like a gap, but this analyst sees it as a strength. Adversarial testing built in-house by a data security vendor tends to be shallow, closer to a v1 feature than a seasoned pen tester. Partnering with an established specialist gives buyers depth and better assurance of a safe AI deployment. Whether this capability is bundled in the license or priced separately was not confirmed.

Email as an AI attack surface most buyers are not yet considering. Picture an AI assistant that briefs its owner each morning on overnight email. An attacker only needs to deliver one message in the inbox, with malicious instructions hidden in an innocuous-looking attachment. The assistant reads it before the owner has arrived at their desk, follows the hidden instructions, and exfiltrates whatever sits in its context window, all inside a routine morning summary. The barrier is near zero: no credentials, no user error, and the victim never sees the payload. Most AI security products do not inspect email at all, because they sit at the AI application or web layer. Forcepoint covers it as a direct extension of its email DLP heritage, closing a gap most organizations are not yet thinking about.

Analyst Perspective

The AI security category is filling up fast. Every data security vendor, every network vendor, and a wave of month-old startups now sells an AI security platform. How is a CISO supposed to navigate this space? The answer depends on what matters most to you: protecting your data, securing agentic AI, or keeping your AI infrastructure free of vulnerabilities. If data leakage is your primary concern, vendors coming from data security have a structural head start.

AI risk is, fundamentally, a data problem: what the model was trained on, what an agent can reach, what leaves in a prompt, what an MCP server returns. A vendor that already classifies data, tracks how it moves, and enforces policy on it is extending a mature engine into a new channel. A vendor starting at the model layer must build that data foundation from scratch, and that takes years. Not every data security vendor arrives from the same corner. Some come up from posture and discovery, knowing what data exists and who can reach it. Forcepoint comes from the enforcement corner: two decades of inline data loss prevention across web, email, endpoint, and network, built to stop data in motion at the moment it tries to leave.

That heritage is what makes the AI agent gateway and the cross-channel shadow AI controls a strong contender in this category. Brokering agent access and blocking a personal-tenant upload are enforcement problems, and enforcement is the harder half of this category to build. Discovery tells you a risk exists; it does not stop it. A platform coming from posture and discovery is still working toward blocking, while Forcepoint is extending blocking it already does across a new set of channels. For AI risk that shows up as data in motion, whether a prompt, an upload, or an agent action, DLP experience is a strong fit.

None of that resolves the harder problem, which is not the tool. It is organizational. The AI footprint arrives faster than any team can govern it, and the security leader is usually the last consulted and the first held accountable. In one recent member conversation, an infrastructure leader wanted a data-protection framework in place before the business scaled AI, while his peers were already building delivery pipelines and project plans on top of tools no one had vetted. He described it as being asked to design the plane while others were already flying it. No platform fixes that misalignment. What a platform can do is give the accountable leader enough visibility and enforcement to catch up. Forcepoint is a credible answer to that need, as long as the buyer understands the tool is the smaller half of the problem. The larger half is aligning the organization on who governs AI before the footprint overruns them. That's not a problem any vendor can solve.

Conclusions and Fit Guidance

Strong Fit

Existing Forcepoint data security customers extending into AI. This is the clearest fit. The AI controls sit on the same platform, policy engine, and AI mesh classification layer already in place, so the buyer extends a known deployment rather than standing up a new tool. The licensing and the enforcement reach both build on what they already run.

Buyers whose AI risk extends beyond sanctioned SaaS. Where sensitive data can reach AI tools over the web, email, endpoint, or personal accounts, not just sanctioned cloud apps, Forcepoint's enforcement sits in those channels already. Its shadow AI controls can distinguish a corporate tenant from a personal one and block the personal path while allowing the sanctioned one. The value tracks directly to how much AI risk lives outside sanctioned SaaS. Buyers whose shadow AI risk is concentrated at the endpoint should confirm current coverage, since endpoint-agent discovery is on the near-term roadmap rather than shipping at GA.

On-premises or air-gapped buyers. Forcepoint's AI controls inherit the data security platform's deployment flexibility, running SaaS, private cloud, on-premises, or fully air-gapped. As much of this category ships SaaS-only, that leaves defense industrial base, government, and other sovereignty-constrained buyers without a path. Forcepoint is a credible option where data-residency rules or air-gap requirements rule out a cloud-only tool.

No-Fit or Requires Augmentation

Buyers who need predictable, fixed billing. Forcepoint's AI security is priced on consumption, weighted to data volume. Buyers are entitled to the full set of enforcement points and charged on what they use. In an environment where AI usage is growing fast and often unpredictably – agent counts and token consumption can climb faster than anyone forecasts – that model can produce a bill that swings month to month.

Buyers whose priority is model integrity and supply-chain assurance. Forcepoint's AI security did not include an AI Bill of Materials or model supply-chain scanning at launch, and the vendor could not confirm it was on the roadmap. Buyers who need to verify model provenance, detect tampering, or scan AI components for vulnerable open-source dependencies should not expect that from Forcepoint today. This is a data-in-motion and agent-governance platform, not a model-integrity one.

Latest Technology Notes

All Technology Notes
Over 100 analysts waiting to take your call right now: +1 (703) 340 1171