Skip to content

Commit caf2618

Browse files
authored
Merge pull request #569 from HighnessAtharva/updates
add browser plugin to release notes
2 parents c3fa4da + 9b7ec8a commit caf2618

3 files changed

Lines changed: 37 additions & 3 deletions

File tree

docs/getting-started/3.5-release.md

Lines changed: 37 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -28,6 +28,8 @@ v3.5 expands AccuKnox across the supply chain, the Rule Engine, AI workloads, an
2828

2929
**AI Security**
3030

31+
- [Prompt Firewall Browser Plugin](#prompt-firewall-browser-plugin-chrome-and-firefox)
32+
- [Red Teaming Custom Models via Collector](#red-teaming-custom-models-via-collector)
3133
- [AI App Inventory: Managed and Unmanaged](#ai-app-inventory-managed-and-unmanaged-assets)
3234
- [AI Widget Builder on Dashboard V3](#ai-widget-builder-on-dashboard-v3)
3335

@@ -89,6 +91,41 @@ Every collector row now carries a **Trigger Scan** action alongside the existing
8991

9092
![Collectors page with Trigger Type column showing Scheduled and On Demand entries](./images/release-notes/v3.5/collectors-page-overview.png)
9193

94+
### Prompt Firewall Browser Plugin: Chrome and Firefox
95+
96+
The AccuKnox Prompt Firewall is now available as a browser extension for Chrome (and Chromium-based browsers — Brave, Edge) and Firefox. The plugin sits between your browser and ChatGPT, intercepting every prompt and response before it travels to or from the model. If a prompt trips an active policy, a red banner appears inline. Prompts that clear all policies pass through silently with a green confirmation.
97+
98+
This gives organizations direct control over how employees interact with public AI assistants, without requiring any application-level SDK integration or proxy changes. Install it, paste your integration token from **AI Security > Integrations**, and policies you have already configured in AccuKnox apply immediately.
99+
100+
![AccuKnox Prompt Firewall extension popup showing connected status and running scan counts](../integrations/image-66.png)
101+
102+
A blocked prompt surfaces a red banner inline in ChatGPT, and an allowed prompt shows a brief green clearance — both without leaving the browser tab.
103+
104+
![Red banner shown in ChatGPT when a prompt is blocked by an active Prompt Firewall policy](../integrations/image-63.png)
105+
106+
![Green banner shown in ChatGPT when a prompt clears all active Prompt Firewall policies](../integrations/image-64.png)
107+
108+
Every intercepted prompt and response is logged under **AI Security > Request Log** with timestamp, direction (PROMPT or RESPONSE), verdict (ALLOW or BLOCK), content preview, and latency. The log is searchable, filterable by verdict or direction, and exportable as JSON.
109+
110+
**Installation guides:**
111+
112+
- [Chrome, Brave, and Edge →](/integrations/openai-browser-integration/)
113+
- [Firefox →](/integrations/openai-firefox-browser-integration/)
114+
115+
!!! note "Firefox installation"
116+
Firefox loads the extension as a temporary add-on via `about:debugging`. It will unload when the browser restarts. Contact AccuKnox support if you need a permanent deployment option for Firefox.
117+
118+
### Red Teaming Custom Models via Collector
119+
120+
AccuKnox supports cloud-based onboarding to scan and red team every model deployed across an AWS Bedrock account. The Collector method is a lighter alternative for targeting a specific model on Bedrock without onboarding the entire cloud account or organization.
121+
122+
- **Targeted scope:** Red team a single Bedrock model instead of every model in the account.
123+
- **No account onboarding needed:** Works directly against the model's inference endpoint using a Bedrock API key.
124+
- **Fast setup:** A single Custom Model collector covers the connection, prompts, and schedule.
125+
- **Flexible:** The same flow works for any LLM exposed over an HTTP endpoint, not just Bedrock.
126+
127+
[Set up red teaming for a custom model →](/how-to/aiml-bedrock-collector/)
128+
92129
### AI App Inventory: Managed and Unmanaged Assets
93130

94131
AI/ML inventory now splits cleanly between **Managed** assets (models hosted by your cloud providers such as Bedrock foundation models) and **Unmanaged** assets (Python and Node packages, local dependencies, and models pulled into arbitrary VMs).
@@ -294,9 +331,6 @@ The following items shipped or partially shipped in v3.5 but are not yet documen
294331
- **SSPM (SaaS Security Posture Management)** — Owner: Ayush. Collector is exposed in the Add Collector flow but no PRD or validation context yet. Need product scope, supported SaaS apps, and screenshots.
295332
- **Nessus Web Host Endpoint Findings** — Owner: Ayush. Web-finding data type added on the Nessus integration; need a tenant with live data, finding detail screenshots, and parser coverage notes.
296333
297-
### AI Security
298-
299-
- **Bedrock and Custom-Model Red Teaming via Collector** — Owner: AI team. The Custom Model option under LLM Red Teaming accepts a Bedrock endpoint URL, but the **Predefined Prompts / Custom** toggle is mislabelled and the documentation of supported endpoints is incomplete. Need the corrected UX and a working end-to-end example.
300334
-->
301335

302336
AccuKnox v3.5 sharpens coverage where security teams spend the most time, pipelines, cloud assets, ticketing hygiene, and AI workloads, while opening the platform to Japanese and Korean speakers.
-5.26 KB
Loading
-6.82 KB
Loading

0 commit comments

Comments
 (0)