Anchor
The compliance model behind ZeroDrift.
Anchor is a purpose-built compliance SLM that checks every AI message against regulations, firm policies, restricted data rules, and customer-specific controls, and rewrites, blocks, or escalates violations in real time.
Architecture
Model:
Anchor 3.0-Preview
Design:
A purpose-built compliance SLM and a deterministic rules engine, run in parallel and reconciled into one verdict
Decisions:
Pass · Rewrite · Block · Escalate
Coverage:
Agents, email, chat, marketing, documents, and APIs
Explainability & Revalidation:
Every rewrite is revalidated before delivery; every verdict cites the exact rule it applied
Modality:
Text
Deployment:
Cloud or in-VPC
Core Proof Points
What makes Anchor different.
Deterministic where it matters. Intelligent where it needs context.
A deterministic rules engine and the Anchor SLM check every message in parallel, reconciled into one verdict.
Every verdict is explainable.
Shows the policy, rule, or regulation behind each decision.
Stopped before delivery.
Anchor rewrites, escalates, or blocks non-compliant communication, and every rewrite is revalidated before it ships.
Tuned to your firm.
Adapts to your policies, restricted lists, approval flows, and voice.
How Anchor Decides
Every decision is grounded in the controls your firm already owns.
Anchor runs every AI-generated message through the same evaluation pipeline, in real time, before delivery.
AI
Output
AI Agents
Chat and Messaging
Marketing
APIs
Documents
Anchor Evaluation
//01
Regulation match
//02
Firm policy match
//03
Restricted data detection
//04
Context and intent analysis
//05
Customer-specific control check
//06
Final verdict and explanation
Pass
Safe output delivered
Rewrite
Compliant version delivered
Block
Stopped before delivery
Escalate
Routed for human review
Example Decision
A verdict your compliance, AI, and security teams can understand.
Original AI Output
Composed {
"This fund is a safe way to increase returns and is a strong fit for your portfolio."
}
Anchor Verdict
Verdict:
REWRITE
Rule Cited:
FINRA Rule 2210
Policy Match:
Promissory performance language
Policy Match:
The output implies safety, guaranteed suitability, and future performance.
ZeroDrift Compliant Rewrite
Sent {
// Rewritten
"This fund may be appropriate for certain investors depending on their objectives, risk tolerance, and time horizon. Past performance does not guarantee future results."
}