A prompt carries more out of the building than an attachment.
For decades, companies protected networks, devices, files and applications. AI opens a new channel: employees transfer business context directly into external models through natural language. Names, customer context, contract details and source code go with it.
Governing what happens before the model sees the data.
An AI Interaction Firewall checks each prompt before processing: it evaluates the content, detects sensitive information, understands intent, applies policy and decides whether the request is allowed, masked, blocked or routed to a local model.
Classic security alone can't secure AI conversations.
Traditional tools were built for structured data flows, file transfers and network traffic. AI prompts are fluid, semantic and context-dependent. A keyword filter misses meaning; a regex sees a phone number but not the confidential business situation. BRANE adds the missing interaction layer.
Five places your data leaks into AI today.
Every prompt is a potential breach. These are the patterns we see across every audit.
- A
PII in employee prompts.
Names, addresses, IDs pasted into ChatGPT — often without realising. Cloud training risk, GDPR risk.
- B
Trade secrets in “draft this for me…”.
Roadmaps, pricing models, M&A drafts — sent to public LLMs as helpful context. No mask, no log.
- C
Patient IDs to external APIs.
Clinical workflows that touch OpenAI or Anthropic without on-prem masking. PHI exposure, audit nightmare.
- D
Source code uploads.
Developers debug with Copilot or Claude. IP leaves the building. No vault, no provenance.
- E
Contract clauses to summarisers.
Legal teams paste NDAs and MSAs into AI summarisers. Terms become training data for foreign models.
Where BRANE sits
Employees
typing into ChatGPT, Gemini, Claude
BRANE
checks every prompt before it leaves the network
Cloud models
the uncritical, masked
Local models
the sensitive, in your own environment
Firewalls protect networks. DLP protects files. BRANE protects what employees type into AI. Because it sits in front, you decide which model is used for what.