Every doc, email,
and web page.
Scanned. Contained.
Your AI reads content from everywhere, and attackers hide instructions inside it. heySec Prompt Firewall scans that content for hidden attacks and locks it inside a secure boundary, so even what we don’t catch can’t give orders to your model.
3–12 ms
Scan and contain in one call
100+ languages
Catches attacks in any language
2 layers
Scan to detect, contain to isolate
Two layers of protection
Every doc, email, and web page.
Scanned and contained.
heySec Prompt Firewall sits between incoming content and your AI. We scan it for hidden attacks and lock it inside a secure boundary your AI treats as data. Customer emails, uploaded files, web pages, support tickets. Your users never see a delay. Your AI never sees the attack, and even unknown tricks stay trapped.
The problem
Your AI is only as safe as the content it reads
Modern AI apps read customer emails, uploaded documents, web pages, and tool outputs, then trust every word of it. Attackers hide instructions inside that content to trick your AI into leaking data, changing its behaviour, or overriding safety rules. Prompt Firewall scans for known attacks and contains the content so even new techniques cannot break out.
Every trick, handled
Attackers are clever. We’re ready.
Attackers hide instructions by disguising text, translating it into other languages, or burying it inside files and images. Our scanning layer strips away every disguise and reads what the attacker really wrote. Our containment layer wraps the content inside a boundary your AI treats as plain data, so even new tricks stay trapped.
Simple to add
Simple API. Flexible by design.
Scanning and containment in the same run. Drop it into any architecture: RAG pipelines, tool-use chains, multi-agent systems, or a simple chatbot. Same API, same protection, whatever pattern you build with.
What we catch
The disguise never reaches your AI
Attackers use every trick they can think of to sneak instructions past your AI. Here are the main ones we stop, in plain English.
Hijack attempts
Content that tries to tell your AI to ignore your rules or pretend to be a different assistant. Caught in 100+ languages.
Scrambled text
Instructions hidden inside codes or scrambled formats that look like gibberish to humans but give orders to your AI. We unscramble every layer.
Invisible characters
Letters that look normal to you but read as something completely different to your AI, plus hidden characters that the human eye cannot see at all.
Hidden messages
Secret instructions tucked inside the first letters of paragraphs, between spaces, or hidden in patterns designed to fool your AI while looking like ordinary text to you.
Impersonation
Fake conversations and pretend authority figures designed to trick your AI into thinking an attacker is you, your company, or a trusted user.
Attacks inside files
PDFs, Word docs, spreadsheets, slide decks, and images. We open every file, read everything inside, and scan it before your AI does.
Why not use another AI to check?
Because the same trick that fools one AI will fool another
If an attacker can manipulate your AI with hidden instructions, they can manipulate a second AI the same way. Running one model in front of another just gives the attacker two targets instead of one, slower responses, and a bigger bill. We go deeper: purpose-built scanning that can’t be talked out of doing its job, plus containment that locks content in a boundary no attack can escape.
100×
faster
Checks complete in 3 to 12 ms. A second AI typically takes 500 ms or more. Your users never notice a delay.
Reliable
The same piece of content always gets the same verdict. No guesswork, no surprises, no retries needed.
Two
layers
Scanning catches known attacks. Containment locks content in a boundary your AI treats as data. Together, they give you defence in depth against prompt injection.
Enterprise edition
Inside your walls.
Under your control.
For government, defence, finance, and healthcare. Prompt Firewall can run fully inside your own network with no outside connection at all. Your data never leaves your premises. Same scanning and containment as the cloud, fully in your hands.
Privacy & trust
Built in Europe. Held to the highest standard.
heySec is an EU company, registered and operated entirely within the EU. Your data never touches a server outside Europe.
EU servers only
All compute, storage, and processing runs inside the European Union. Your content never crosses an EU border.
Every verdict explainable
Every request returns a verdict with detected threat categories, severity, and confidence. Deterministic and auditable, ready for incident review or compliance reporting.
GDPR-compliant by design
Personal data in logs is automatically detected and redacted before storage. Built to meet GDPR Arts. 5, 25, and 32.
Make your AI safe to read the real world
Join the private beta. One API call gives you scanning and containment from day one.