SentinelOne brings Prompt Security to AWS Sydney for AI sovereignty

SentinelOne has made its Prompt Security product available through Amazon Web Services' Sydney region, giving Australian organisations a local data-residency option for AI security monitoring.

The rollout addresses a growing concern for businesses and public sector bodies using tools that inspect employee prompts, model responses and autonomous agent activity, because those services may themselves process sensitive information such as customer records, source code and credentials.

Australian business use of AI has risen sharply, according to SentinelOne, while regulators have stepped up warnings about the risks that come with broader adoption. APRA has identified prompt injection, data leakage, insecure integrations and misuse of autonomous agents as emerging issues for regulated entities.

For organisations in financial services, insurance, government and critical infrastructure, the physical location of data processed by security services has become a procurement and compliance issue. Hosting requirements and sector risk frameworks have increased scrutiny of technology providers handling operational data generated through AI tools.

SentinelOne described the Sydney deployment as among the first of its kind in the local market. The product is designed to discover, govern and secure AI systems used by employees, developers and autonomous agents, including the information shared with those systems and the external tools agents are allowed to access.

That reflects a shift in how companies approach AI controls. Early efforts often focused on whether staff were using approved chatbots, but the spread of AI agents and AI-enabled software has widened the issue to include what these systems can access, what they can send out, and where records of those exchanges are stored.

Jason Duerden, Area Vice President, Australia and New Zealand at SentinelOne, linked the launch to the pace of local uptake.

"AI has moved into the daily operations of Australian organisations faster than most security and governance programs were built to accommodate. As of July, Australia ranked number one on Anthropic's Claude usage index, and security must be prioritised to support this rapid adoption," said Jason Duerden, Area Vice President, Australia and New Zealand at SentinelOne.

He also pointed to the effect of data-location rules on security buying decisions.

"For government, financial services, insurance and critical infrastructure, data location can determine whether an AI security service is viable, particularly when it may inspect employee prompts, customer data, proprietary information and autonomous-agent activity. Local hosting removes a major barrier, while SentinelOne Prompt Security gives organisations the visibility and controls to stop sensitive information leaving through AI and set clear boundaries around what agents can access and do," said Jason Duerden.

Partner uptake

Regional channel partners are already building services around the Sydney-hosted version. AU Cyber in Australia and Advantage in New Zealand are among the first to invest in related offerings, suggesting service providers see demand from customers that need local hosting alongside AI monitoring and policy enforcement.

That interest comes as advisers and managed service providers look for new work tied to AI governance. Many customers now need help assessing where prompts are stored, which models are used inside business processes, and how agent-based tools connect to internal systems.

Three areas

SentinelOne groups the product into three functions. One covers employee and developer use of AI tools, including discovery of sanctioned and unsanctioned services and application of policies intended to prevent leakage of sensitive material through prompts.

A second focuses on AI applications themselves. The product tests for prompt injection, jailbreaks and data poisoning, then applies runtime controls intended to block malicious prompts and unsafe outputs.

The third covers agentic AI. In this area, the service is designed to discover AI agents, tools and Model Context Protocol servers, enforce least-privilege access and create an audit trail of agent actions and interactions with business systems.

Australian authorities and intelligence partners have increasingly warned about those same issues. Guidance from the Australian Signals Directorate and its Five Eyes counterparts has urged organisations to use continuous monitoring, strict privilege controls and strong identity management for agentic AI, particularly when agents can interact with connected tools outside traditional security boundaries.

Itamar Golan, General Manager of SentinelOne's Prompt Security, said the risk lies in the interaction itself.

"AI security has to operate where the risk occurs, in the live exchange between a person, model, agent and the tools it can call. Prompts can contain customer records, financial data, source code and credentials, while agents can hold persistent access to enterprise systems. Prompt Security gives security teams real-time visibility and enforcement across those interactions, allowing organisations to use AI productively without giving employees or autonomous agents unchecked access to sensitive information," said Itamar Golan, General Manager of SentinelOne's Prompt Security.

Prompt Security became part of SentinelOne after an acquisition in 2025 and now sits within the company's wider AI security portfolio, which covers employees, applications, infrastructure and autonomous agents.