The service is readily available and integrates seamlessly with standard tools.
Haven't used AI Cordon yet?
Ask Trustburn
Questions about AI Cordon
What is AI Cordon?
AI Cordon is a software service that provides prompt injection detection for applications built on large language models (LLMs). It inspects documents, RAG context, tool outputs, and user messages for hidden instructions, returning a verdict in milliseconds before the input reaches a model or agent. The service runs its own reference-free detector called IPI, which does not rely on an external LLM in the detection loop. Users can access the product through an inspect interface and an API, with a free tier of 1,000 checks.
How does the IPI detector work?
The company publishes measured performance metrics for its detector, including false positive rate, true positive rate, and ROC AUC. Some reviewers say the service effectively identifies hidden commands in documents before they reach the AI system. The platform allows users to block injections with their cursor or use the API for automated protection.
What are the pricing and free tier limits?
The service offers a free tier of 1,000 checks and a paid plan starting at $40 per million documents. Users can start free with up to 50,000 documents. The system flags documents with a probability and the specific line that triggered the alert for human review.
How do I access the service?
The platform provides an inspect interface and an API for integration. Users can access the product through these two methods. The API returns a verdict, probability, and the triggering line for each document.
What are the key benefits of using AI Cordon?
The service claims fewer than 1 false alarm in 10,000 clean documents. It screens every outside document for hidden instructions before the AI sees them. This helps protect systems from indirect prompt injections in emails, web pages, and uploaded files.
How does the semantic detector differ from keyword matching?
The service reads what the model understands, not just the words. A compact open model reads the document once, and the detector reads how the model understands it to see if it sees a command. This approach catches hidden instructions that might be missed by simple word matching.
What types of documents can be inspected?
The service can be used to protect email, web, and uploaded files from hidden commands. It also inspects tool or MCP output and READMEs with setup lines. The goal is to prevent hidden instructions from acting on the AI system.
What did the company's tests reveal about indirect prompt injections?
The service claims to have tested against 12 agent models and found that 8 of them would file a payment based on a planted line in an ordinary business email. It also found that 6 of 10 agent models would run a command to install a tool for themselves. These tests demonstrate the potential risks of indirect prompt injections.
How does the API response help with review efficiency?
The service provides an API that returns a verdict, probability, and the triggering line for each document. This allows users to review one line instead of the whole email. The API is designed for fast inspection with a verdict in milliseconds.
What are the performance claims and onboarding steps?
The service claims to have fewer than 1 false alarm in 10,000 clean documents. It also provides a free tier of 1,000 checks. The platform is designed to be easy to start with a free tier and scale up to paid plans.
How can I contact AI Cordon?
You can reach AI Cordon by e-mail at [email protected].
What is the official website of AI Cordon?
The official website of AI Cordon is ai-cordon.com.
How is AI Cordon rated on Trustburn?
On Trustburn, AI Cordon has a rating of 4.8 out of 5 based on 4 reviews.
Post your question publicly — other visitors, and whoever manages this page, can answer. It will be visible on this page.
Want to hear about this company again? Follow it and we’ll email you when a new review appears.
Based on reviews, listing data and the company's website. Verify anything important with the company directly.
4 Reviews
Getting started with AI Cordon was a breeze. I just signed up, grabbed an API token, and made a few test calls. The pay-as-you-go model is super convenient since I only pay for what I use, with no subscription fees hanging over my head. The free 1,000 checks gave me plenty of room to experiment without any risk. I tried it with a few RAG chunks and documents, and the integration into my existing pipeline was smooth. The documentation is clear, and the examples in Python and JavaScript were really helpful. It feels like a solid tool for catching those sneaky prompt injections before they mess with the model. I like that it handles up to 5,000 characters per document, which covers most of my use cases. The fact that it runs a compact open model internally means it's fast and doesn't rely on external services for every check. It's great to have a simple, effective solution that just works.
The range of detection capabilities is impressive, covering RAG chunks, tool outputs, and documents with a single semantic detector. The ability to run a compact open model in one pass without an external LLM loop suggests low latency and cost efficiency. With metrics like an 80.3% catch rate at a 0.01% false positive rate, the solution appears robust for various injection types. The clear distinction between intent and keywords adds a layer of security that traditional keyword matching might miss. Additionally, the support for multiple programming languages and frameworks like LangChain makes integration straightforward. The evidence-based approach, with quantified performance metrics and confidence intervals, provides transparency. This level of detail helps developers trust the detector's verdicts. The option to scan entire knowledge bases in one run is particularly useful for large datasets. Overall, the product offers a comprehensive and secure way to protect LLM applications from prompt injection attacks.
The pay per check model looks sensible for early adoption, but I wonder if volume discounts apply to larger teams?
Questions & Answers
No questions yet — be the first to ask.
Sign in to post
Your text is saved — you'll come right back to post.
or with email
Don't have an account? Create one
News & Press
No news yet. This is where the business shares its own updates and press.
Photos
No photos yet. Photos shared by reviewers will show up here.
About AI Cordon
AI Cordon is a software service that provides prompt injection detection for applications built on large language models (LLMs). It inspects documents, RAG (retrieval-augmented generation) context, tool outputs, and user messages for hidden instructions, returning a verdict in milliseconds before the input reaches a model or agent. The service runs its own reference-free detector called IPI, which does not rely on an external LLM in the detection loop. Users can access the product through an inspect interface and an API, with a free tier of 1,000 checks. The company publishes measured performance metrics for its detector, including false positive rate, true positive rate, and ROC AUC, based on a benchmark of over 200,000 documents.
- Website
- ai-cordon.com
- [email protected]
