Threat protection that runs itself.
Rules catch what you already know. The AssurAI model, trained in-house on governance and attack patterns, runs in-line at the gateway. Monitoring agents built on it watch every thread of AI activity and bring your team only the decisions that need a person.
What the AssurAI model detects.
Prompt injection
Direct and indirect injection, including instructions hidden in documents, web pages and tool results.
Jailbreak attempts
Patterns that try to push a model outside its policy, in chat interfaces and in agents.
Data exfiltration
Personal data, credentials, payment details and source code leaving through prompts, responses or connectors.
Unsafe tool use
Agent actions outside approved scope: bulk exports, payments, deletions, unexpected recipients.
Governance classification
Every prompt and response classified against your policies and data classes.
Risk signal
What the model sees feeds the continuous AI risk score, so the score reflects real behaviour.
Observe. Evaluate. Surface.
Observe
Agents follow every thread of activity through the gateway: prompts, responses, tool calls and connector traffic.
Evaluate
The model classifies each event against your policies and known attack patterns, and separates routine traffic from misuse.
Surface
Routine traffic is logged and threats are blocked. The few cases that need judgement reach the right owner with the evidence attached.
What your team sees.
On a typical day: a few held actions to approve, one or two incidents to acknowledge, and a weekly summary.
- Response agents
- Block, hold for approval, revoke abused access, open and close routine incidents.
- Integration
- Alerts and records export to your SIEM, ticketing and chat tools.
- Tuning
- Adapted to your estate and policies. Your data is never used to train for other customers.
- Residency
- Runs in your environment or a dedicated EU tenant.
- Scope
- Defensive only. AssurAI tests systems only with their owners' consent and does not conduct or sell offensive capability.
Frequently asked.
Does this replace our security operations centre?
No. It adds the AI layer and feeds your existing SOC and SIEM with AI-specific detections and evidence.
What happens when the model is unsure?
The action is held and routed to the owner with the context attached. Nothing risky executes silently.
Is our data used to train the model for others?
No. Tuning on your estate stays with your deployment.