Bedrock Data extends ArgusAI governance to Google Vertex AI

Bedrock Data has added governance support for AI agents built on Google Vertex AI Search and Dialogueflow, extending its ArgusAI product to another major cloud AI platform.

The integration allows security and governance teams to identify what data a Google AI agent can access and apply access rules to that data. It covers applications built on Vertex AI Search, Dialogueflow CX and Dialogueflow ES.

ArgusAI already supports AI systems running on Amazon Bedrock, Snowflake Cortex AI and ChatGPT Enterprise. With the addition of Google Vertex AI, Bedrock Data can apply the same policy model across several widely used AI environments instead of requiring teams to set separate governance rules for each platform.

Google Coverage

The new support addresses a problem facing companies that have rolled out AI agents across internal and customer-facing systems. Tools such as Dialogueflow are often used in customer service, while Vertex AI Search is used to build applications that retrieve information from internal data sources. In both cases, security teams need to know what sensitive information those systems may expose in response to a user query.

ArgusAI maps the data connected to each Google AI application and classifies it by sensitivity, data type and business domain. This lets teams see not only which records or sources an agent can access, but also what kinds of information could surface through it, including personal data, financial information and security-related data.

It automatically discovers Vertex AI Search and Dialogueflow applications and maps the data stores they access, including Google Cloud Storage buckets. It then builds what Bedrock Data calls a Data Bill of Materials, linking each agent to the underlying data sources it uses.

That inventory is designed to update continuously as agent configurations or connected data change. According to Bedrock Data, this provides a basis for applying governance controls to AI systems in the same way organisations track software components or cloud assets.

Policy Model

A central part of the product is the ability for teams to write data access policies in plain English. Those policies can then be enforced automatically, with alerts generated if changes to agent settings or underlying data create a breach.

Examples include rules that prevent any Dialogueflow agent from accessing personally identifiable information, or restrictions that allow only a named agent to reach items such as Social Security numbers or licence numbers.

The system is built on Bedrock Data's Metadata Lake, which the company describes as a graph-based knowledge base that maps enterprise data across sensitivity, lineage, entitlements and business context. With the new integration, that graph now extends to Google AI agents as well.

"Enterprises are accelerating AI adoption to stay competitive, and their proprietary data is their competitive advantage. Bedrock Data ArgusAI is built to facilitate scaling AI by enabling safe access to their most proprietary datasets so AI can produce business outcomes," said Bruno Kurtic, Chief Executive Officer and co-founder of Bedrock Data. "Today we extend ArgusAI to Google Vertex AI, one of the leading AI platforms. ArgusAI now governs enterprise AI across Amazon Bedrock, Snowflake Cortex AI, ChatGPT Enterprise and Google Vertex AI."

Cross-Platform Rules

AI governance becomes more difficult as enterprises adopt multiple AI platforms, each with its own structure for data access. In practice, different services refer to underlying data links in different ways, which can force security teams to rebuild the same policy logic several times.

Bedrock Data argues that a common governance layer avoids that duplication. By representing each AI system and its connected data in a single graph, it aims to let a policy written once apply across Google, Amazon and Snowflake environments without changing the policy model itself.

Pranava Adduri, Chief Technology Officer and co-founder of Bedrock Data, described that approach. "Each AI platform defines data access in its own way. Vertex AI Search has data stores, Amazon Bedrock has knowledge bases and Snowflake Cortex has Cortex Search. Governing them separately means rebuilding policy logic for each one," said Adduri. "Instead, we represent every AI system and the data it touches in one graph inside the Metadata Lake. A policy written once applies consistently whether the agent runs on Google, AWS or Snowflake. Adding Vertex AI meant extending the graph connections. The governance model didn't change."

ArgusAI for Google Vertex AI Search and Dialogueflow is available now within the Bedrock Data platform.