Enterprise AI Accountability

Every AI action.
Logged, signed,
provable.

Vigil is infrastructure that makes AI agents accountable. Cryptographic audit trails, real-time enforcement, and automated damage recovery — across every AI provider your enterprise uses.

Read the API docs
8
Detection layers per request
<50ms
Pre-request gate latency
VOAF
Open audit standard
vigil — live interception stream
LIVE
3 agents · 1 blocked · 0 exfiltrated
$4.88M
Average enterprise data breach cost in 2024. AI agents expand the attack surface.
40%
Of enterprise applications will embed AI agents by end of 2026. — Gartner
Zero
AI providers offer cross-provider accountability. Structural conflict of interest.
18 mo
Estimated window before EU AI Act and MAS mandates require agent audit trails.
How it works

A proxy that sits between your users and every AI they trust

Vigil intercepts every AI API call in real time. It decomposes semantics, runs 8-layer anomaly detection, enforces your policy rules, and writes a tamper-evident audit log — before the response reaches the user.

Demo: Audit trail

Court-admissible proof of every AI action

Vigil writes every AI interaction to a VOAF-format audit trail. Each record is SHA-256 hashed and chain-linked to the previous entry. The chain is verifiable and tamper-evident. Click any record to inspect it.

  • 01Every interaction logged within the proxy, before response delivery
  • 02Hash chain links records — modification of any entry breaks the chain
  • 03VOAF open standard export readable by third-party auditors
  • 04Jurisdiction-ready: EU AI Act, MAS TRM, Colorado SB 24-205
VOAF Audit Log — vigil-core v0.9.1
record_idv-4471-001
timestamp2026-03-19T10:47:24Z
provideranthropic
actionchat.completion
anomaly_score0.08
statusallowed
sha256_hasha3f2...8d91
record_idv-4471-002
timestamp2026-03-19T10:47:26Z
provideropenai
actiontool.call[read:email]
anomaly_score0.61
statusflagged
sha256_hash7c1e...2f44
record_idv-4471-003
timestamp2026-03-19T10:47:29Z
provideropenai
actionchat.completion
anomaly_score0.97
statusBLOCKED
sha256_hashe90b...3a12
Hash chain integrity
block[001]: a3f2...8d91
block[002]: 7c1e...2f44 ← prev: a3f2...8d91 ✓
block[003]: e90b...3a12 ← prev: 7c1e...2f44 ✓
Demo: Kill switch

Revoke any agent's authority in under one second

When an agent behaves outside its authorized scope — or when you simply decide to stop it — Vigil revokes all active sessions, OAuth tokens, and API credentials atomically. No partial state. No lingering access.

  • 01One-click revocation across all active sessions and OAuth grants
  • 02Atomic: all permissions revoked in a single operation, not sequentially
  • 03Evidence package generated automatically on kill switch trigger
  • 04Dead man's switch: authority expires without active attestation
Authority ControlAll agents authorized
AGT
financial-assistant
Scope: read:portfolio, execute:trades
AUTHORIZED
AGT
legal-research-bot
Scope: read:docs, search:case-law
AUTHORIZED
AGT
customer-comms-agent
Scope: read:crm, send:email
AUTHORIZED
Atomically revokes all agent authorities and generates evidence package
Demo: REPAIR mode

Undo what a compromised agent did

When Vigil detects a policy violation after the fact, REPAIR mode reconstructs what the agent did and reverses it. OAuth tokens revoked. Transactions cancelled. Evidence packaged for legal or insurance use.

  • 01Action rollback within configurable time window
  • 02OAuth token revocation across connected services
  • 03Transaction cancellation hooks for financial integrations
  • 04Evidence package: VOAF export, signed incident report, timeline reconstruction
vigil-repair — incident #4471
Use cases

Built for industries where AI mistakes have consequences

FS
Financial services and family offices

AI agents with portfolio access, trading authority, and payment execution need cryptographic accountability. A compromised financial agent is an existential event. Vigil provides the kill switch, rollback, and evidence trail that institutional risk desks require.

  • Kill switch on any agent holding financial authority
  • Transaction rollback within configurable window
  • VOAF evidence packages for dispute resolution
  • MAS TRM and MiFID II audit trail compliance
  • Dead man's switch: authority expires without attestation
LC
Legal and compliance

Law firms and compliance teams deploying AI for research, contract drafting, and regulatory filing need proof of what the AI produced and what data it accessed. Vigil's VOAF trail is court-admissible and jurisdiction-aware.

  • Immutable log of every document the AI read or produced
  • Prompt injection detection with flagged evidence preservation
  • EU AI Act, SOC 2, and GDPR reporting templates
  • Behavioral drift alerts when AI recommendations shift
  • Attorney-client privilege boundary enforcement
IN
Insurance platforms (B2B2C)

Embed Vigil into your product as an accountability layer for policyholders using AI agents. The audit trail becomes a claims defense asset. The kill switch becomes a product differentiator. One API integration. Per-seat pricing.

  • White-label audit trail for your end customers
  • Verified AI behavior logs for claims dispute resolution
  • Partner API with per-seat billing
  • Behavioral baseline as underwriting signal
  • Automated incident reports for claims processing
W3
Crypto and Web3

Agents executing on-chain transactions operate with zero human oversight. Vigil enforces authorization boundaries, flags anomalous transaction patterns, and revokes signing authority before damage compounds. Compatible with existing wallet infrastructure.

  • Transaction signing authority enforcement
  • Anomaly detection on transaction value and frequency
  • Kill switch stops agent before irreversible on-chain execution
  • VARP: Vigil Agent Revocation Protocol for smart contract integration
  • Collective threat signatures across the Vigil network
Product architecture

Four operating modes that stack as your risk profile grows

Start with prevention. Add recovery. Add persistence. Add offense. Each mode builds on the last. Your subscription tier determines which modes activate.

01
SHIELD
Guardian tier and above

Pre-request blocking gate. Runs synchronously under 50ms. 8 detection layers. Stops threats before they reach your AI.

  • Prompt injection detection
  • Jailbreak pattern matching
  • Context poisoning analysis
  • Authority boundary enforcement
  • Policy rule engine
02
REPAIR
Fortress tier and above

Post-incident recovery. Rolls back agent actions, revokes tokens, cancels transactions, and generates signed evidence packages.

  • Action rollback engine
  • OAuth revocation
  • Transaction cancellation
  • Evidence package generator
  • Incident timeline reconstruction
03
SENTINEL
Citadel tier

Persistent threat monitoring. Adversary fingerprinting. Behavioral drift detection. Legal pipeline automation. Runs continuously in background.

  • Persistent threat monitor
  • Adversary fingerprinting
  • Behavioral baseline drift
  • Legal automation pipeline
  • Collective threat feed
04
WARDEN
Add-on · Coming 2026

Offensive countermeasures. Active adversary disruption. Network-scale collective defense using aggregated threat signatures from across the Vigil deployment base.

  • Passive intelligence (L0)
  • Active countermeasures (L1)
  • Offensive response (L2)
  • Collective defense (L3)
  • Cross-network threat sharing
Regulatory credibility

Built to the emerging global AI governance standard

Vigil submitted a formal response to NIST Docket 2025-0035 on AI accountability infrastructure. Comment tracking number: mmk-190r-hvap.

Submission filed March 2026

Vigil's architecture aligns with IMDA's Model AI Governance Framework for agentic AI. SHIELD, REPAIR, and SENTINEL map directly to the framework's accountability requirements.

Architecture aligned

Automated regulatory report generation for three jurisdictions. VOAF export format is readable by third-party auditors without proprietary tooling.

Report templates live

An open standard for AI agent audit logs. Published as a standalone spec at github.com/vigilsec/voaf-spec. Not tied to Vigil's product code.

v1.0.0 published

Protocol for cryptographic attestation of AI agent authority grants. Enables verifiable "the human authorized this" assertions that survive audit.

Spec in development

Open protocol for cross-system agent authority revocation. Enables kill switch propagation across third-party systems and smart contracts.

Spec in development

Open standards published independently at vigilsec.ai — not tied to Vigil product code.

B2B2C integration

Embed accountability into your product in one day

Vigil's Partner API lets you offer your users cryptographic AI audit trails, kill switch controls, and behavioral monitoring as a native feature of your platform. One integration. Per-seat pricing. Full white-label.

Python — Vigil SDK
# Wrap any AI client in one line
from vigil_sdk import VIgilClient

# OpenAI — drop-in replacement
client = VIgilClient(
  api_key="your-vigil-key",
  underlying="openai",
  policy="strict"
)

# Every call is now monitored
response = client.chat.completions.create(
  model="gpt-4o",
  messages=[{"role": "user", "content": prompt}]
)

# Access the audit trail
audit = client.audit.export_voaf(
  format="signed",
  jurisdiction="sg-mas-trm"
)
Partner API — enroll a user
# POST /v1/partner/users/enroll
{
  "user_id":    "your-platform-user-id",
  "email":      "user@yourplatform.com",
  "tier":       "fortress",
  "policy":     "financial-services-strict"
}

# Response: signed license key + audit endpoint
{
  "license_key": "vig_live_...",
  "audit_url":   "https://vigil-cloud.fly.dev/...",
  "modes":       ["shield", "repair"]
}

What your users get inside your product

You don't send them to runvigil.ai. The accountability layer lives inside your product, branded as yours. Vigil is the infrastructure.

AI activity monitorLIVE
LAST 24 HOURS
1,247
REQUESTS
3
BLOCKED
12
FLAGGED
AGENT AUTHORITIES
financial-assistantACTIVE
research-agentACTIVE
Get access

Deploy today. No meeting required.

Self-serve signup for Guardian, Fortress, and Citadel. Enterprise and partner integrations get a response within 24 hours.

Enterprise and partner inquiry

The AI agent accountability layer your enterprise is missing.

No single AI provider can solve this. It requires a cross-provider infrastructure layer. That is what Vigil is.

Read the API docs
Local-first · No AI data leaves your device · VOAF open standard · NIST aligned