Netskope to integrate AI Guardrails with Amazon Bedrock AgentCore

Netskope has announced a forthcoming integration between its One AI Guardrails product and Amazon Bedrock AgentCore, AWS's platform for building, connecting and scaling AI agents in enterprise environments. The integration is designed to extend Netskope's existing security controls into agentic workflows, where autonomous software agents act without direct human input.

The companies say the integration will allow Netskope's detection layer to feed signals into AgentCore's policy engine, which enforces controls at the gateway rather than inside the agent's reasoning loop. Netskope describes this as a meaningful architectural distinction: detection can be probabilistic, drawing on machine-learning models to identify threats, but enforcement remains deterministic, producing a consistent allow-or-deny decision on every agent action.

What the integration covers

Once available, the joint capability is intended to cover several threat vectors specific to agentic AI systems. Netskope One AI Guardrails will contribute prompt injection detection, sensitive data exposure protection, toxic output filtering, restricted topic enforcement and model response validation. AgentCore then acts on those signals across agent-to-tool, agent-to-LLM and agent-to-agent communications.

John Martin, Chief Product Officer at Netskope, framed the announcement around a broader shift in enterprise risk: "The newest software in your enterprise doesn't wait for a human to click, and it demands a new security model. The agentic era requires the same security rigor for software agents that we've always applied to human users."

Netskope did not disclose a general availability date, pricing, or which specific AWS regions will be supported at launch. The press release carries an explicit forward-looking statement noting that functionality and timing remain subject to change and should not be relied upon in purchasing decisions.

Market context and competitive positioning

The security of agentic AI pipelines is a rapidly developing sub-category within the broader AI security market. As enterprises move from AI experimentation to production deployments, the attack surface expands significantly: agents can call external APIs, read and write to datastores, and chain actions across third-party tools, creating vectors for prompt injection, data exfiltration and privilege escalation that traditional CASB and SASE controls were not built to address.

Netskope sits in a competitive field that includes Palo Alto Networks (with its AI Access Security offering), Zscaler's AI security controls, and a growing number of specialist startups addressing LLM and agent security. Hyperscalers are also building native guardrail capabilities; Amazon's own Bedrock Guardrails product provides a baseline layer, making the decision to partner rather than compete a notable positioning choice for Netskope.

Netskope notes existing collaborations with Anthropic's Project Glasswing and OpenAI's Trusted Access for Cyber programme, which broaden its positioning as a vendor aligned with frontier model providers rather than treating them as adversaries in the enterprise stack.

Regulatory read-across

Enterprise customers deploying AI agents in regulated sectors will be watching this space closely. The EU AI Act classifies certain autonomous systems as high-risk, and the act's obligations around human oversight and logging of consequential decisions are directly relevant to agentic architectures. In the UK, the ICO has indicated that automated decision-making under UK GDPR requires documented controls, and the FCA has signalled scrutiny of AI use in financial services workflows. US federal guidance from NIST's AI Risk Management Framework also emphasises the need for monitoring and logging at inference time. Deterministic gateway-level enforcement of the kind Netskope describes fits well with the audit-trail requirements these frameworks are beginning to codify, though buyers will want to see independent validation of detection accuracy rates before relying on the controls for compliance purposes.