Shadow AI Just Got Visible: Vocus and Fortinet Launch SASE Control for ChatGPT, Gemini, and Claude

TL;DR

Vocus and Fortinet launched Vocus Secure Shield, a managed SASE platform designed to provide enterprise visibility and control over employee use of LLM applications (ChatGPT, Gemini, Claude). The announcement signals two critical shifts: (1) Shadow AI has become a board-level governance problem, and (2) the SASE layer is becoming the control plane for agentic AI.

What Happened

On May 3, 2026, Vocus (Australian managed services provider) and Fortinet (network security vendor) announced Vocus Secure Shield—a fully managed Unified Secure Access Service Edge (SASE) platform built explicitly to close the "shadow AI visibility gap."

The product is positioned to address a specific governance challenge: employees using public LLM applications (ChatGPT, Google Gemini, Claude) with enterprise data while IT has no visibility into what data is being shared, which model processes it, or where it's stored. Vocus Secure Shield aims to provide:

• Real-time visibility into LLM app usage across the workforce
• Data classification and sensitive content enforcement before it reaches an LLM
• Policy controls to block or regulate use of specific AI models
• Compliance logging for audit and regulatory requirements

This is a managed service play—customers don't buy a point product; they outsource shadow AI governance to Vocus's SOC + Fortinet's SASE backbone.

Why This Matters for Enterprise Defense

The Shadow AI Governance Gap Is Now Critical Infrastructure

Two years ago, "shadow AI" meant a few engineers experimenting with ChatGPT. Today, it means systematic organizational risk:

• Credential leakage: Developers pasting Slack tokens, API keys, and database connection strings into ChatGPT for debugging
• Sensitive data exfiltration: Product roadmaps, customer PII, financial data, and source code flowing into LLM training pipelines
• Compliance violations: HIPAA-covered data, PCI-DSS secrets, and classified information processed by third-party models without consent
• Prompt injection attack surface: Enterprise AI agents trained on poisoned data sourced from public LLMs

SASE Is Becoming the AI Policy Enforcement Layer

The move by Vocus/Fortinet isn't isolated. It reflects a broader industry shift: network control planes (SASE, SD-WAN, Zero Trust) are evolving into AI governance platforms. Why?

• SASE sits at the edge — every application request (including API calls to ChatGPT, Gemini, Claude) passes through it
• Policy is declarative — once a rule is deployed, it applies uniformly across thousands of endpoints
• Data classification flows through — DLP (Data Loss Prevention) engines can scan payloads before they leave the network
• Compliance auditing is native — SASE logs everything, making it audit-ready for regulatory frameworks

This is functionally equivalent to how cloud access security brokers (CASBs) evolved in 2015-2020. SASE is the new CASB, but for direct API consumption of third-party models.

The Underlying Threat Model

Vocus/Fortinet's play assumes enterprise environments are now managing two classes of AI risk:

1. Unsanctioned external AI (ChatGPT, Gemini, Claude) — uncontrolled by the organization

2. Sanctioned internal AI (enterprise models, Copilots, agents) — controlled but still dangerous

The gap between them is where breaches happen. A developer uses ChatGPT to debug a private API endpoint, leaks the secret, and now an attacker has direct access. Or a financial analyst copies a spreadsheet with YoY projections into Claude, and the data influences the model's weights—which affects all future users of that model.

Vocus Secure Shield tries to close that gap by making the SASE layer aware of which traffic is destined for external LLMs and applying data policies before it leaves.

Lyrie Assessment: The Autonomous Defense Angle

This announcement reflects a critical reality for Lyrie.ai's mission around autonomous defense:

Enterprise AI Governance Is Now a Speed Game

Enterprises have maybe 18-24 months before agentic AI becomes embedded in their operations (Microsoft Copilot in Teams/Office, Salesforce Agentforce, OpenAI Swarms). By then, shadow AI won't be a compliance problem—it will be a supply chain risk.

If your developers are training their AI agents on data that came from unvetted LLM outputs, your agents inherit the poisoning. This is the new attack surface: AI training data provenance.

SASE + Autonomous Policy Enforcement = The New Defensive Boundary

Vocus Secure Shield is fundamentally a manual policy engine — an admin writes rules, SASE enforces them. But as attack velocity accelerates, manual policy becomes a liability. The next generation of SASE will be agentic:

• AI agents autonomously detect data classification anomalies (e.g., "this looks like source code but it's being posted to ChatGPT API")
• Autonomous policy generation (e.g., "Block all API calls to GPT-4 from users in Finance department when payload contains credit card data")
• Real-time model-switching (e.g., detect that Claude is drifting off-policy and automatically route requests to an on-premise model instead)

Vocus/Fortinet will either have to ship autonomous policy layers or lose to vendors who do.

This Is a Market Signal That Enterprise AI Governance Is Consolidating

Vocus partnering with Fortinet signals that shadow AI visibility is no longer a nice-to-have. When established infrastructure vendors (Fortinet) are willing to integrate third-party AI governance, it means board-level demand is real. This will accelerate:

• Consolidation of point solutions (standalone shadow AI tools) into SASE platforms
• Regulatory mandates for AI transparency (NIS2 Phase 2, EU AI Act enforcement in late 2026)
• Insurance requirements for enterprises to demonstrate AI governance or face premium hikes

Recommended Actions for CISOs

Immediate (This Week)

1. Audit shadow AI usage — deploy a SASE or cloud DLP solution to measure how much of your egress traffic is going to public LLMs

2. Classify sensitive data — identify which data classes (PII, credentials, source code, financial) are at risk if leaked to external models

3. Draft an AI usage policy — define which LLM services are allowed, for which roles, and under what conditions

Short-term (This Month)

1. Evaluate SASE solutions with shadow AI visibility — Vocus Secure Shield, CrowdStrike Falcon Cloud, Zscaler Private Edge, Fortinet FortiGate Cloud are starting to ship this capability

2. Integrate with your CI/CD pipeline — block commits that contain secrets, then extend to block commits that attempt to push code to a development LLM for analysis

3. Run threat modeling on agentic workflows — if you're planning to deploy autonomous agents (GitHub Copilot in CI/CD, Salesforce Agentforce for customer service), model where poisoned data can enter the training loop

Long-term (Strategic)

1. Treat AI supply chain like software supply chain — SBOMs for training data, attestation of model versions, audit trails for model updates

2. Invest in anomaly detection at the SASE layer — manual policies scale logarithmically; you need autonomous policy to keep pace with threat velocity

3. Prepare for AI governance insurance requirements — soon, enterprises won't be able to insure their AI operations without proving governance controls exist

Sources

1. Vocus News: Vocus and Fortinet launch Secure Shield — Published May 3, 2026

2. Telecompaper: Vocus and Fortinet launch managed SASE platform targeting 'shadow AI' risks — Published May 4, 2026

---

Lyrie.ai Cyber Research Division