MNPI Filter

The mnpi-filter policy detects material non-public information (MNPI) signals in AI interactions to prevent insider trading compliance violations under SEC Rule 10b-5, Regulation FD, and MAR (EU Market Abuse Regulation). It screens for patterns indicative of unannounced earnings, pending mergers, regulatory actions, and material contract details that have not been publicly disclosed. Because MNPI violations carry criminal liability, this policy defaults to hard blocking.

Use this page when

• You need to prevent insider trading compliance violations by detecting material non-public information in AI interactions.
• You are deploying AI in investment banking, hedge funds, or corporate finance where MNPI exposure creates legal liability.
• You want to enforce Chinese wall information barriers in AI tools used by advisory and trading teams.

Primary audience

• Primary: AI Agents, Technical Engineers
• Secondary: Technical Leaders

Configuration

pack:
  name: mnpi-filter
  version: "1.0.0"
  enabled: true

policies:
  chain:
    - mnpi-filter

policy:
  mnpi-filter:
    detect_patterns:
      - "quarterly earnings will be"
      - "merger announcement before"
      - "pending acquisition of"
      - "undisclosed revenue figures"
      - "board approved the deal"
    action: block

Fields

Field	Type	Description	Default
detect_patterns	string[]	MNPI risk patterns to detect. When empty, Keeptrusts uses built-in defaults that detect earnings previews ("Q[1-4] earnings will exceed"), merger/acquisition discussions ("pending acquisition", "merger target"), regulatory action mentions ("SEC investigation", "enforcement action pending"), material contract details ("contract value not yet disclosed"), and insider knowledge signals ("before the announcement", "not public yet"). Custom entries are merged with the built-in list.	[]
action	enum: "block"	Action to take on detection. Always block — sharing MNPI through AI tools constitutes a securities law violation regardless of intent.	"block"

Use Cases

Investment Banking Chinese Wall Enforcement

Prevent deal team members from leaking MNPI into AI tools that could cross information barriers (Chinese walls) between advisory and trading desks.

pack:
  name: mnpi-filter
  version: "1.0.0"
  enabled: true

policies:
  chain:
    - mnpi-filter

policy:
  mnpi-filter:
    detect_patterns:
      - "deal valuation is"
      - "target company accepted"
      - "fairness opinion draft"
      - "synergy estimate confidential"
      - "break-up fee terms"
      - "pre-announcement trading"
      - "pitch book for acquisition"
    action: block

Hedge Fund Compliance

Screen AI interactions for sell-side analysts and portfolio managers to prevent MNPI from influencing trading decisions through AI-assisted analysis.

pack:
  name: mnpi-filter
  version: "1.0.0"
  enabled: true

policies:
  chain:
    - mnpi-filter

policy:
  mnpi-filter:
    detect_patterns:
      - "channel check indicates revenue"
      - "supply chain data suggests earnings"
      - "insider confirmed the guidance"
      - "management indicated privately"
      - "non-public order book data"
      - "pre-release clinical trial results"
    action: block

Corporate Treasury AI

Prevent treasury and finance teams from sharing material financial data with AI tools before public disclosure — protecting against Regulation FD violations.

pack:
  name: mnpi-filter
  version: "1.0.0"
  enabled: true

policies:
  chain:
    - mnpi-filter

policy:
  mnpi-filter:
    detect_patterns:
      - "cash position as of"
      - "debt covenant breach"
      - "dividend decision before announcement"
      - "share buyback authorization pending"
      - "credit facility renegotiation"
      - "material weakness in internal controls"
    action: block

Earnings Season Lockout with Full Finance Stack

Combine with financial-compliance and pii-detector for comprehensive financial services AI governance, especially during quarterly quiet periods.

pack:
  name: mnpi-filter
  version: "1.0.0"
  enabled: true

policies:
  chain:
    - mnpi-filter
    - financial-compliance
    - pii-detector
    - audit-logger

policy:
  mnpi-filter:
    detect_patterns:
      - "actual EPS before release"
      - "revenue miss not yet reported"
      - "guidance revision embargoed"
      - "restatement pending"
    action: block

  pii-detector:
    action: redact

  audit-logger:
    immutable: true
    retention_days: 2555
    log_all_access: true

How It Works

1. Pattern scanning — Incoming prompts and outgoing model responses are scanned against the MNPI pattern list (built-in defaults plus custom entries). Patterns are matched using case-insensitive substring and regex matching.
2. Contextual signal detection — The filter looks for co-occurring signals that indicate MNPI context: temporal indicators ("before the announcement", "not yet public"), materiality markers ("material", "significant impact"), and confidentiality signals ("confidential", "restricted", "embargoed").
3. Blocking — Any pattern match triggers an immediate block. The response is replaced with a compliance notice referencing the MNPI policy violation. The original content is never forwarded to the model or returned to the user.
4. Audit trail — Every block generates a structured audit event including the matched pattern, timestamp, user identity, and policy name. These events are available through the Keeptrusts events API for compliance reporting.

Combining With Other Policies

Policy	Combined Effect
financial-compliance	Adds broader financial regulation patterns (KYC/AML red flags, wash trading signals). Together with mnpi-filter, covers both securities law and general financial compliance.
pii-detector	Redacts personal identifiers (account numbers, SSNs) that may accompany MNPI in financial contexts.
safety-filter	Adds a broader content-safety backstop around financial conversations that may drift into prohibited territory.
human-oversight	Escalates borderline high-risk turns for compliance review rather than relying only on automated blocking.
audit-logger	Ensures all interactions, not just blocked ones, are logged for regulatory examination readiness.
data-routing-policy	Routes financial AI traffic only to providers with appropriate data handling certifications (SOC 2 Type II, ISO 27001).

Best Practices

• Start with built-in defaults — The default patterns cover the most common MNPI categories (earnings, M&A, regulatory actions, material contracts). Add custom patterns specific to your firm's deal pipeline and business lines.
• Enforce during quiet periods — Tighten MNPI patterns during quarterly earnings quiet periods, blackout windows, and active deal processes. Consider maintaining separate policy configs for normal and restricted periods.
• Layer with information barrier controls — MNPI filters are a technical control, not a substitute for organizational Chinese walls. Use alongside access controls that restrict which teams can use which AI configurations.
• Never downgrade to warn — MNPI violations under SEC Rule 10b-5 carry penalties up to $5M and 20 years imprisonment for individuals. The action field only accepts block by design.
• Review patterns quarterly — Update patterns to reflect new deal types, financial products, and regulatory guidance. Remove patterns that generate excessive false positives after thorough compliance review.
• Train users on MNPI boundaries — Technical controls catch known patterns, but novel MNPI formulations may evade detection. Combine with user training on what constitutes MNPI.

For AI systems

• Canonical terms: Keeptrusts, mnpi-filter, detect_patterns, action, block, SEC Rule 10b-5, Regulation FD, MAR, insider trading, material non-public information
• Config/command names: mnpi-filter policy, detect_patterns, action: block
• Best next pages: Financial Compliance, DLP Filter, Human Oversight

For engineers

• Prerequisites: Work with your compliance team to define MNPI risk patterns specific to your business (earnings, M&A, regulatory actions). Built-in defaults cover common patterns.
• Validation: Test with MNPI-indicative phrases and verify hard blocking. There is no redact option — MNPI violations are securities law violations regardless of intent.
• Key commands: kt policy lint, kt policy test, kt events tail

For leaders

• Governance: MNPI violations carry criminal liability under SEC Rule 10b-5, civil penalties up to $5M for individuals, and unlimited disgorgement. This policy defaults to hard blocking.
• Cost: Local pattern matching with no external cost. A single insider trading enforcement action costs orders of magnitude more than AI governance infrastructure.
• Rollout: Deploy immediately for all AI tools accessible to deal teams, research analysts, and corporate insiders. No audit-only mode — securities law compliance requires enforcement.

Next steps

• Financial Compliance — Broader financial controls
• DLP Filter — Data loss prevention
• Human Oversight — Escalation for borderline cases
• Audit Logger — Immutable audit trail for compliance