Agentic AI Security: Thales Expands Google Cloud Partnership

What happened

Thales has widened its relationship with Google Cloud to target a problem many enterprises are only starting to confront: how to secure AI agents that act on their own rather than simply answer questions. The French defense and security group announced on September 28, 2026 that it is integrating its AI Security Fabric with Google Cloud's Gemini Enterprise. The goal is to give organizations protection, visibility, and policy enforcement over AI-driven workflows running on Google's platform.13

The integration covers interactions among users, agents, models, and tools, and applies security and governance controls in real time.13 According to one account, the Fabric is designed to enforce policy, restrict what data agents can reach, and block AI actions that have not been authorized.3 A market-focused summary described the arrangement as protecting AI agents, data, and interactions across the full AI lifecycle.2

The partnership was also presented as one of five announcements at the Thales Cyber Summit in Paris on October 1, 2026.4 The other four were a hardware security module for post-quantum protection, an AI-enabled data security posture management (DSPM) product, software protection aimed at AI-powered attacks, and a new global cybersecurity approach built around real-time threat detection and response.4

Why agents change the security equation

Thales framed the deal around a shift in how companies deploy AI. Eva Rudin, Senior Vice President for Cybersecurity Products at Thales, said enterprises are moving from AI assistants to AI agents that can take actions, make decisions, and connect to critical business systems on their own. She argued this calls for a fundamentally different approach to security.1

That framing holds up. A chatbot that drafts text has a limited blast radius. If it is manipulated, the damage is usually confined to a bad answer. An agent with access to tools, data stores, and other agents is a different kind of risk. If it is fed malicious instructions hidden in a document or web page, or if it trusts output from another agent that has been compromised, it can take real actions with real consequences. These are the broad categories security researchers group under prompt injection and agent-to-agent risk.

The announcements do not spell out specific defenses against those named attack types. Still, the capabilities described line up with them. Real-time visibility across user, agent, model, and tool interactions gives defenders a view into the chain of calls where manipulation could happen.13 Limiting data access and blocking unauthorized actions are the controls that would contain an agent once it has been steered off course.3

Where the accounts converge and diverge

The coverage agrees on the core facts. All of it describes an expanded collaboration centered on AI Security Fabric and agentic workflows on Google Cloud.1234 The differences are mostly about emphasis.

• The corporate release focuses on trust and governance as the foundation for enterprise AI adoption.1
• Trade coverage stresses compliance and secure scaling. It is also the most concrete about enforcement functions such as limiting data access and blocking actions.3
• The market summary describes the deal as a straightforward partnership covering the AI lifecycle.2
• The summit report treats it as one item in a broader portfolio responding to increasingly automated, AI-driven threats.4

There is also a small difference in timing. The collaboration was announced on September 28, and the summit packaging came on October 1.14 That suggests Thales announced the Google news first and then folded it into a larger strategic story.

Reading the move

The most useful way to read this deal is as a bet on where security controls will live in agentic systems. Thales is not building the agent platform. It is placing its policy and enforcement layer inside Google's enterprise agent environment. This mirrors earlier cloud security history, when third-party vendors embedded themselves in hyperscaler ecosystems instead of competing with them.

The broader summit lineup supports that reading. Post-quantum hardware, data posture management, and software protection are all areas where Thales already has credibility.4 Agentic AI governance is newer ground. Partnering with Google gives Thales distribution and relevance in a category where standards and buyer expectations are still forming.

What remains unclear from the announcements is how well these controls perform against determined attackers. That includes how granular the policies can be and whether they cover agents that span multiple vendors. Enterprises evaluating the integration should ask for specifics on detecting injected instructions and on governing agent-to-agent trust. These are the areas where agent security is most likely to be tested.