Secureline product / 02
Promptgate
Between your users
and your model.
Promptgate is part of Secureline's product collection. Secureline has not yet published its feature list, so this page explains the problem a prompt-security layer addresses and marks clearly what is still to be confirmed.
Enquire about a Promptgate demo01 / The problem
A prompt is
an input you don't control.
An application that sends user text, documents or web pages to a language model is passing untrusted input to a system that follows instructions written in plain language. The model cannot reliably tell your instructions from an attacker's.
Two risks, in OWASP's words
Prompt injection. OWASP's Top 10 for LLM Applications lists prompt injection first, and notes that techniques like retrieval-augmented generation and fine-tuning do not fully mitigate it. Content from external sources can alter the model's behaviour in unintended ways, and it says it is unclear whether fool-proof prevention exists.[1]
Sensitive information disclosure. LLMs embedded in applications risk exposing personal data, financial details, health records, confidential business data and credentials through their output. OWASP notes that rules placed in the system prompt may not always be honoured and can be bypassed by prompt injection.[1]
02 / Where a prompt-security layer sits
Checked on the way in.
Checked on the way out.
The common pattern is a control point between the application and the model. It looks at what is going in and what is coming back, and applies a policy before either side acts on it.
- RequestUser or app sends a promptChat box, API call, or a document the model will read
- Check inInspect the inputInjection patterns, secrets, personal data, policy limits
- ModelModel answersAny LLM, hosted or self-run
- Check outInspect the outputLeaked data, unsafe content, unexpected instructions
- DeliverApplication receives a responseAllowed, redacted or blocked, with a log entry
03 / What such a layer does
The jobs,
and the limits.
These are the measures OWASP describes for these two risks. They are the jobs a prompt-security layer is generally built to do. Which of them Promptgate does is listed as pending below.[1]
- Screen input. Apply semantic filters and string-checking to scan for non-allowed content before it reaches the model.
- Redact sensitive data. Use tokenization and pattern matching to detect and redact confidential content before processing.
- Check output. Evaluate responses (for example for relevance and groundedness) to spot potentially malicious or leaking output.
- Limit privilege. Enforce least-privilege access, so a manipulated model can reach less.
The honest limit. OWASP says measures like these can mitigate the impact of prompt injection, not eliminate it. A prompt-security layer is one control among several; it does not replace permission design, testing and monitoring.[1]
04 / Pending confirmation
What we can't
tell you yet.
Promptgate is a Secureline product with a dedicated identity. The details below have not been confirmed by the product owner, so this page does not state them.
- What Promptgate checks
- Pending product-owner confirmation
- How it is deployed (proxy, API, SDK)
- Pending product-owner confirmation
- Supported models and integrations
- Pending product-owner confirmation
- Data handling and retention
- Pending product-owner confirmation
- Canonical product URL
- Pending product-owner confirmation
- Next step
- A product-specific enquiry →
For a useful first conversation
What would you need to see?
Bring your AI use case: which model, what data it can reach, and what a bad outcome would look like. This link opens the local contact page with Promptgate selected; it does not launch a simulated product.