ADX Vision 2.0 Data Sheet: See the AI tools your people use. Govern what goes into them.

ADX Vision 2.0

ADX Vision 2.0 is an independent control layer for AI. It maps AI usage across your organization from the endpoint and applies your policy to AI interactions rather than trusting the AI system to decide.

On the endpoint. LLM redirection. Prompt protection.

Your AI risk isn’t the tools. It’s what goes into them.

Employees paste intellectual property, customer records, and strategy docs into AI tools every day. It leaves through a text box, one prompt at a time, past every control built for files and networks.

• 47% of security leaders have had sensitive data exposed via AI.
• 77% of those with an incident said a sanctioned AI tool was involved.

Source: BlackFog/Sapio 2026

See first. Then govern.

ADX Vision maps AI tools in use and governs the data flowing in, at the moment of input. Policy on the data, not the tool, so the employee keeps ChatGPT and the customer record stays behind.

1. Inspect: Inspects data moving to AI tools in real-time, including inside encrypted browser sessions and apps.
2. Control: Applies policy independently of the model or service, separating sanctioned AI from Shadow AI by behavior, not a list of domains.
3. Block: Blocks sensitive transfers in real-time, before data leaves the device. Sanctioned AI keeps moving.

Why ADX Vision

• On the endpoint. Rules out cloud proxy players.
• Not just the browser. Rules out browser & extension tools.
• Across major AI tools. Rules out vendor-native admin panels.
• Without blocking the tool. Rules out the block-everything approach.

How ADX Vision protects

01. Sensitive prompt interception

Sensitive data stops before it reaches the AI tool.
Intercepts intellectual property, customer records, credentials and financial data at the moment of input, before it’s sent to a third-party AI tool.

02. AI activity monitoring

Shadow AI stops being a guess. It becomes an inventory.
See the AI tools in use across your organization, on every device, including the ones nobody sanctioned.

03. Prompt injection protection (7 layers)

Data doesn’t only leave by mistake.
Compromised pages, poisoned context, and rogue agents can manipulate an AI session into pulling data out. Manipulated prompts are identified before they create risk.

04. Custom policies

Your definition of sensitive, not ours.
Custom controls apply at company or departmental level, from credit card and Social Security numbers to the terminology that matters to your business.

05. LLM redirection

Send people to the AI you actually licensed.
AI requests can be redirected to sanctioned, officially licensed LLM services, so employees keep working in the services you approved.

06. Data rights

Know what each AI tool does with your data.
ADX Vision tracks the license terms behind each model: retention windows, whether your prompts train the model, and whether traffic is encrypted.

07. Prompt audit trail

An optional record of every prompt, ready before anyone asks.
Audit prompts sent to frontier LLMs and retain them under the policy you set, for governance, compliance and audits.

08. Usage & cost analytics

Token spend across all major models, in one view.
A unified view of token consumption across LLMs, showing where tokens are consumed and by whom. It also makes token theft visible.

Don’t ask the AI to police itself.

Agentic AI:

• On the endpoint. Policies apply at the endpoint, so AI interactions are governed consistently across different LLMs and services.
• Visibility and auditability. One view across endpoints and AI interactions, covering what is used, what is sent and what it costs.
• Data stays inside. Whatever an agent is asked to do, sensitive information is held back before it leaves the business.

Beyond the network.

Reads the data inside encrypted browser sessions, where CASB and SWG see only a trusted domain. Covers AI tools including ChatGPT, Claude, Gemini, and Perplexity, not only sanctioned enterprise deployments. Managed and unmanaged devices, personal machines, offline and air-gapped, even autonomous agents.

AI created a new security gap. ADX closes it.

ADX prevents exfiltration at the moment of AI interaction. DLP, EDR, DSPM and other security categories were built to solve different problems. ADX adds the control they weren’t designed to provide.