Vaikora › Blog › Product Updates
How Vaikora enforces email-based threat signals at the AI agent layer before an action runs
A phishing link that bypasses your email filter is a problem. The same link arriving as context inside an AI agent prompt, triggering a tool call to an external API, is a different category of problem entirely. Standard email security controls stop at the inbox. They have no visibility into what an agent does with the content it receives.
This article explains how Cyren's email threat intelligence and Vaikora's runtime enforcement work together, and what that combination actually prevents.
Where email security normally ends
Cyren Email Security for Microsoft 365 connects to your tenant through Microsoft Graph. There is no MX record change, no mail rerouting. It runs analysis inside a virtual machine in your own Azure subscription and reports into your Sentinel workspace.
The control it applies is at delivery time: a URL is flagged, a file is quarantined, a sender is blocked. That is the right place to intervene for a human user reading their inbox.
The gap opens when AI agents enter the picture. An agent might receive a forwarded message, a parsed email summary from another tool, or a user instruction that includes a URL the agent will then act on. At that point, the email security layer has already done its job. It evaluated the message on arrival. It did not evaluate what the agent will do with the content three tool calls later.
What Cyren GlobalView telemetry contributes
Cyren's GlobalView network processes over a billion email, URL and file transactions per day. That scale produces one specific operational advantage: Cyren detects phishing and malicious infrastructure during staging, before campaigns are weaponized. The indicators it generates reflect what is being assembled right now, not what was confirmed malicious last week.
Those indicators include:
- Domains registered or repurposed for phishing
- URLs that lead to credential-harvesting infrastructure
- File hashes tied to active malware distribution
Cyren pushes these indicators into CrowdStrike Falcon and SentinelOne. Inside a Sentinel deployment, Cyren ships as a packaged solution through the Microsoft Sentinel Content Hub, including a connector, analytic rules, hunting queries, and workbooks. The indicators are available as structured telemetry, not just as alerts requiring human review.
Where Vaikora picks up
Vaikora sits inline between an AI agent and every LLM or external tool it calls. Before each action executes, Vaikora evaluates the action against deterministic policy and returns one of four decisions: allow, block, modify, or require approval. The P50 enforcement latency is 8ms. The P99 is 45ms. Policy runs at 10,000 or more actions per second.
The integration with Cyren threat intelligence feeds directly into that evaluation. If an agent is about to call an external URL, make an API request to a domain, or pass content to an external service, Vaikora checks the target against current threat intelligence before the call leaves the system.
A concrete scenario: a user asks an AI agent to summarize a document and send it to a webhook URL. The agent constructs the tool call. Vaikora inspects the destination URL against Cyren's indicator feed. The domain is flagged as phishing infrastructure registered three days ago. Vaikora blocks the tool call before it executes. The agent never reaches the external service. The decision is written to a SHA-256 hash-chained, tamper-evident audit log, with the reason recorded.
That audit record is what a compliance review or incident investigation needs. Not a firewall log. A timestamped, signed record showing which agent, which action, which policy matched, and what was blocked.
The gap between detection and enforcement
Detection tells you something happened. Enforcement stops it before it does. The common failure mode in AI agent deployments is that organizations apply detection controls designed for human workflows and assume they cover agent behavior. They do not. An agent does not pause to read a security alert. It executes.
Vaikora integrates natively with Microsoft Sentinel, Microsoft Defender, SentinelOne, CrowdStrike Falcon, and AWS Security Hub. High-severity AI agent threats are promoted into Azure Security Center and Microsoft Defender for Cloud as native incidents, with no custom engineering. Threats are also pushed into SentinelOne as Threat Intelligence indicators with severity mappings: Critical at 100, High at 75, Medium at 50, Low at 25.
The result is that email threat intelligence, which starts as a signal about delivery-time risk, becomes an enforcement guardrail inside the AI agent runtime.
What to check in your environment
If your organization is running AI agents that process email content, URL inputs, or user-supplied external references, review whether your current threat intelligence is reaching the agent execution layer, not just your inbox filter and your SIEM.
Specifically, check:
- Whether URLs extracted from email content are evaluated against current threat intelligence before an agent acts on them
- Whether your threat intelligence indicators are flowing from Cyren or your SIEM into a runtime control point, or only into alerting workflows
- Whether agent tool calls to external services produce a signed, auditable record that maps to specific threat intelligence matches
If the answer to any of those is "we don't know", that is the right starting point for a gap assessment.
See Vaikora enforce policy on your AI
Open-core AI runtime control. Self-host the MIT gateway free, or run the hosted Control Plane.
Get a demo Self-host the gateway
Vaikora