Compare
Lakera guards an AI application against attacks and leaks. Hushbridge protects the client details a person puts into a prompt.
Different job: AI application guardrails
Lakera is now part of Check Point. The deal was announced in September 2025 and closed on 22 October 2025. Its docs now call the product Check Point AI Guardrails. It sits in front of a model or agent your team runs. It looks for prompt attacks, unsafe tool use and data leaks, then blocks or masks. It fits security and engineering teams who ship their own AI apps.
Local protection layer
Hushbridge works where people already use AI. On your device it swaps details for placeholders such as [PERSON_1] and shows you what the AI will get. After you review it, the placeholder version goes to the AI. Real values come back on your device. It fits professionals using ChatGPT or Claude for client work, and developers who want that step in their own code.
| Question | Lakera | Hushbridge |
|---|---|---|
| What it protects | The AI application. It screens inputs and outputs for prompt injection, jailbreaks and data leakage. Its PII detection lists eight types: personal names, US mailing addresses, US phone numbers, email addresses, IP addresses, credit card numbers, IBANs and US Social Security numbers. You can add custom types. | Names, identifiers, addresses, secrets and anything on your own "always protect" list, in text you type or paste. |
| Where processing happens | On the hosted Lakera API by default. With an Enterprise license it can run on-premises or in a private cloud. | On your device. |
| Where original values are stored | Not verified. Its docs say detected PII is masked before it is logged in Check Point systems. | In a swap table that stays on your device. |
| Restore in replies | Not verified. The docs we read describe block or mask. They do not describe putting original values back. | Yes. Real values return on your device in supported replies. |
| How you use it | Your application calls its API on each input and output, then acts on the result. | A browser extension for ChatGPT, Claude, Gemini, Copilot and Perplexity, a web workspace, or the SDK and local proxy in your own code. |
| Deployment | Hosted service, or a self-hosted container run with Docker or a Kubernetes Helm chart. Self-hosting needs an Enterprise license. | Extension version 0.1.1 installs manually in Chrome with "Load unpacked". Workspace and billing run on tightlipai.com during the move to the new name. |
| Pricing | Not published. The site offers a "Start for free" sign-up. Paid terms are not shown. | USD 0.05 per credit. Top-ups from USD 100. USD 20 free credit on sign-up (to be reconfirmed before launch). One credit covers each started block of 8,000 tokens. |
Lakera statements come from its own docs and site. We did not test the product.