Ask FactSet Copilot

Workspace

Overview
Knowledge Graph
Copilot

Knowledge Base

Documents
Ingestion
Metadata & Tags

Governance

Agents
48
Validation Gate
2
Compliance
7
Connectors
Audit Trail
Access Control

Guardian

Guardian Compliance
2
Runtime Alignment
2
Runtime Security

Sovereign posture

Single-tenant VPC

Grounding coverage98.7%

Review controls

Global Markets
Runtime Security

Material UX

Command menu

Search and navigate

Production· eu-central-1

AI red teaming

Offensive datasets replayed against the deployed Guardians — the same policy and security vectors used to fine-tune them. Misses feed straight back into the build pipeline.

Runtime security

Attack vectors666

Detection rate97.0%

Probe suites4

Classes exercised8/10

Probe suites

Each suite replays curated adversarial vectors against a set of OWASP agentic threat classes

Injection & jailbreak sweep

ASI01ASI06

PRB-11 · last run Jul 19

233/240 · 97.1%attention

Tool misuse & privilege escalation

ASI02ASI03

PRB-12 · last run Jul 18

176/180 · 97.8%ok

A2A protocol abuse

ASI07ASI10

PRB-13 · last run Jul 16

90/96 · 93.8%attention

Exfiltration & disclosure chains

ASI05ASI09

PRB-14 · last run Jul 15

147/150 · 98.0%ok

Probes run against a shadow deployment — production traffic is never used as attack input. Every probe result is hash-chained into the audit trail, and misses generate offensive-dataset entries for the next Guardian fine-tuning cycle.