Security & Compliance · AI data firewall
MaskWall
Use AI. Keep your secrets.
A protective layer between your company and every AI service. Keys, passwords and customer data are blocked or masked before they leave, and quietly restored in the reply.
$ ai-wall scan prompt.txt
REDACT SEC016 Password in a connection URL
REDACT SEC001 AWS access key ID
REDACT PII001 Email address
--- what Claude would see ---
Connection: postgres://app:[[DB_PASSWORD_002b3e]]@[[INTERNAL_HOST_4974bc]]/orders
AWS key [[AWS_KEY_dd784f]]. Customer Asha Rao, [[EMAIL_e5e7c6]]
The problem
People paste API keys, database passwords and customer records into AI chats every day. Security teams can't see it, and can't prove to auditors or regulators (GDPR, CCPA, HIPAA, PCI DSS, DPDP) what left.
What it does
Gateway for apps using the Claude API: one setting, ANTHROPIC_BASE_URL, and nothing else changes
Also protects OpenAI, Azure OpenAI, Gemini, Mistral, Groq, OpenRouter and local models through their OpenAI-compatible API, embeddings included
Finds keys, passwords, tokens and private keys, plus emails, phone numbers, cards, IBANs, and national IDs from the US, UK, EU, Australia, Canada and India, verified with their official checksums
Your own code names, client names and internal domains
Blocks, or replaces each value with a placeholder and puts the real value back in Claude's reply, streaming included
The company holds the only Anthropic key; staff get gateway keys with model limits and monthly budgets
Tamper-evident audit log of every request that never stores the values, with CSV export for auditors
Admin dashboard: keys, budgets, spend, what was caught, recent activity
Alerts to Slack or Teams when something is blocked, a budget runs out or customer data appears
Matches your own customer list exactly (only keyed hashes are kept), so it can block your customers' data specifically
Catches prompt injection in documents and tool results, including text hidden in invisible characters
Flags CONFIDENTIAL and "Internal use only" documents; rate limits per key
Claude Code hook: no .env or key files read, no secrets in prompts or code
Chrome extension for claude.ai, ChatGPT, Gemini and Copilot, with settings pushed by IT
Runs on your own server; Docker image included
How it works
- Install the gateway and write your policy
- Point apps at it; install the hook and the extension
- Read the monthly report: what was caught, by whom, and the spend
Works with
Runs in your CI
# Claude Code on every developer machine
ai-wall install-hook
# Apps
export ANTHROPIC_BASE_URL=https://ai-wall.acme.internal
export ANTHROPIC_API_KEY=aw_...
Your data
- Runs inside your network; the company's AI keys never leave the gateway.
- The audit log keeps rule ids and keyed fingerprints, never the sensitive values themselves.
- Policies per team, a dry-run monitor mode, and exact matching against your own customer list.
Who it's for
Questions
Will people notice it?
Rarely. Masked values are restored in the reply, streaming included, so answers read normally. People only see a message when something is blocked.
Does it work with Claude Code?
Yes. A hook stops prompts and tool calls that carry secrets, and keeps .env and key files out of reach.
Can we try it without enforcing?
Yes. Monitor mode finds and logs everything for a week without changing a single request.
Related
Often used with MaskWall
PrivacyPilot
DPDP privacy compliancePrivacy compliance, made practical. Find the gaps between your website and India's DPDP Act, and get the privacy notice, data inventory and consent banner to close them. Notices are checked against GDPR, UK, US and Australian rules too.
₹2,999 per siteView product → Security & ComplianceCERT-In Ready
CERT-In compliance checkCERT-In, checked properly. Check your infrastructure against the CERT-In Directions as written: 180 days of logs held in India, synchronised time and incident reporting.
₹25,000–75,000View product → Security & ComplianceAgentLedger
AI agent inventorySee every AI agent in your code. A complete inventory of the AI agents that can touch your repositories: CI agents, MCP servers, agent settings and organisation-level AI apps.
₹50,000–2,00,000View product →See MaskWall on your own data
A 30-minute call. We run the product on your export, repository or document and walk you through the results.