Vaikora › Blog › Product Updates
Cyren email threat detection inside Vaikora: how pre-campaign signals reach your AI agent guardrails
Phishing infrastructure gets built days before anyone sends a message. The domain is registered, the SSL certificate issued, the hosting provisioned. By the time a user clicks a link, the campaign has been live long enough to have a history. Cyren detects that infrastructure during staging, and Vaikora can act on those signals before an AI agent ever calls the URL.
This article explains how the two systems connect, what happens at the enforcement layer, and what you should verify in your own setup.
Why pre-campaign detection matters for AI agents
AI agents that handle email, ticket queues, or document workflows often call external URLs as part of normal operation. An agent that summarizes incoming email, for example, might fetch a linked resource to give context. If that URL resolves to a phishing page, the agent does not evaluate intent. It fetches the content and continues.
Standard email filters screen messages before delivery. They do not screen tool calls made by an agent at runtime. That is the gap.
Cyren runs on the GlobalView telemetry network, which processes over a billion email, URL, and file transactions a day. That volume produces early signals on newly staged infrastructure because the same domains and IPs appear across multiple detection surfaces before any single campaign fires at scale.
How Cyren feeds indicators into the enforcement layer
Cyren pushes indicators into SentinelOne and CrowdStrike Falcon as first-party threat intelligence. It also ships as a packaged solution in the Microsoft Sentinel Content Hub, including a CCF connector, analytic rules, hunting queries, and workbooks.
Vaikora connects natively to SentinelOne, CrowdStrike Falcon, and Microsoft Sentinel. When Cyren pushes a malicious URL or domain indicator into any of those platforms, Vaikora can consume that signal and enforce against it in real time.
The enforcement path works like this:
- Cyren flags a domain as phishing infrastructure during staging
- The indicator is pushed into SentinelOne or Sentinel as a threat intelligence entry
- Vaikora pulls the updated indicator set and writes a policy that blocks any agent action calling that domain
- When an AI agent attempts a tool call to that URL, Vaikora evaluates the action before execution and returns a block decision
- The block is written to the SHA-256 hash-chained audit log with a timestamp, the matched indicator, and the policy that triggered the decision
The agent never reaches the destination. The decision is deterministic: the same URL evaluated under the same policy always produces the same outcome.
What the Vaikora side actually enforces
Vaikora sits inline between an AI agent and the tools or LLMs it calls. Every action passes through a policy check before execution. The possible outcomes are allow, block, modify, or require approval. For a URL flagged by Cyren as malicious, the standard policy is a block.
Vaikora pushes AI agent threats into SentinelOne as Threat Intelligence indicators with severity mapping: Critical=100, High=75, Medium=50, Low=25. For Microsoft environments, Vaikora promotes high-severity risk into Azure Security Center and Microsoft Defender for Cloud as native incidents without custom engineering.
This matters for security operations teams because the alert lands in the platform they already monitor. No separate queue, no separate tool.
Enforcement latency is P50 8ms and P99 45ms, with a throughput ceiling above 10,000 actions per second. For most agent workflows that is below the threshold where users or downstream systems notice a delay.
What to check in your environment
If you are running AI agents that process email or fetch external content, verify these things:
- Indicator flow: confirm that Cyren indicators are reaching your Sentinel workspace or your SentinelOne/Falcon tenant. The Sentinel Content Hub package includes workbooks you can use to validate throughput.
- Vaikora policy scope: check that your URL and domain inspection policy covers outbound tool calls, not just inbound content. The two are different policy surfaces in most configurations.
- Audit log review: pull the Vaikora audit log for blocked tool calls over the past 30 days and cross-reference against Cyren indicator timestamps. If blocks are appearing before the campaign fired at scale, the pre-campaign detection path is working.
- Severity routing: if you are using the SentinelOne integration, verify that Critical and High severity decisions are creating alerts in your SOC queue and not silently accumulating in a feed no one watches.
The Vaikora Guard MCP open-source package (MIT licensed, published on PyPI as `vaikora-guard-mcp`) is a reasonable starting point if you want to test the enforcement path against a live indicator set before committing to the full control plane deployment.
Start with a single agent workflow that handles external URLs, wire in the Cyren indicator feed, and confirm that a known-bad domain produces a block decision in the audit log before moving to broader policy rollout.
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