AWS freezes Bedrock Agents Classic for new customers

What changed: Amazon has renamed its Bedrock Agents service to Bedrock Agents Classic and, as of July 30, 2026, closed it to new customers, while allowlisted accounts keep full access with no end‑of‑life date announced. The change blocks only two API calls for non‑allowlisted accounts—CreateAgent and InvokeInlineAgent—and freezes the Classic model catalog as of July 30, 2026, meaning no newly released foundation models will appear inside the Classic orchestration layer. AgentCore remains a separate, framework‑agnostic runtime with dedicated services for runtime, gateway, memory, identity, and observability, positioned as the path forward for new production agents.

Why it matters: Teams that already rely on Bedrock Agents Classic can keep their current agents running indefinitely, but they will not see new models inside the Classic console, which will limit long‑term innovation on that stack. New projects should assume AgentCore is the default for building agents on AWS, reducing future migration risk and aligning with where AWS is investing its orchestration capabilities.

Try/watch: Audit which workloads still depend on Classic, document the agents and models in use, and begin testing equivalent flows on AgentCore so you are ready if AWS later introduces stronger deprecation milestones.

Oracle brings Google Gemini into AI Agent Studio

What changed: Oracle is adding Google's Gemini models, including Gemini 3.1 Flash Lite and Gemini 3.5 Flash, to Oracle AI Agent Studio for Fusion Applications and NetSuite, expanding its cloud AI partnership. These Gemini models will sit alongside existing options from Cohere and Meta, giving customers a richer multimodal menu for building Fusion‑native AI agents and embedded AI workflows.

Why it matters: Oracle SaaS customers can now prototype and deploy agents that tap Google's latest models without leaving the Oracle stack, potentially speeding up projects like automated ERP workflows, finance agents, and HR assistants. Solution builders gain a practical way to A/B test different model families for agent tasks—such as complex approvals or document analysis—inside a single orchestration surface rather than stitching together multiple vendors.

Try/watch: For any new agent use case in Fusion or NetSuite, design experiments that compare Gemini against your current models on real business tasks, and track whether multimodal Gemini inputs reduce manual data entry or approval friction.

Cequence extends agentic zero‑trust controls for enterprise AI agents

What changed: Cequence Security has released four new AI Gateway capabilities—AI Discovery, API Registry, LLM Registry, and Skill Registry—along with upgraded Agent Personas that bind an agent’s job description directly to its model, tools, and guardrails. AI Discovery surfaces every agent, AI provider, and Model Context Protocol (MCP) server already running across the enterprise from existing security logs, while the API Registry lets agents call approved APIs without ever holding raw credentials.

Why it matters: Security and platform teams get a way to find shadow agents already in production, standardize how those agents access APIs, and enforce least‑privilege policies without requiring business users to understand security tooling. The enhanced Agent Personas model lets non‑technical stakeholders define what an agent should do and which data it can touch, while AI Gateway enforces those rules automatically, shrinking the gap between AI experimentation and compliant operations.

Try/watch: Use AI Discovery to inventory every agent and MCP server in your environment, then start migrating high‑risk agents onto API Registry and Agent Personas so their access patterns are governed consistently.

Bedrock Data launches Agent DLP for autonomous AI agents

What changed: Bedrock Data has introduced Agent DLP, a runtime data loss prevention capability designed specifically for autonomous AI agents and integrated into its ArgusAI platform. The software sits inline at the agent gateway, bidirectionally inspecting MCP tool calls and responses in real time, and integrates natively with agent gateways such as AWS AgentCore and LiteLLM.

Why it matters: CISOs and data security teams gain a control plane that can actually see and regulate what agents send to external tools and what sensitive data comes back, instead of relying solely on static policies or legacy web proxies. This makes it more realistic to approve high‑value agent workflows—like autonomous report generation or third‑party system updates—without losing visibility into how confidential data moves across tools.

Try/watch: Start by mapping which agents currently have access to sensitive datasets, then pilot Agent DLP on a small set of workflows to test how well its inline inspections catch risky tool calls without breaking business processes.

Crogl offers a free autonomous AI SOC agent for security teams

What changed: Crogl has made its Enterprise AI SOC Agent—an autonomous security operations agent—available as a free download that security teams can deploy within minutes. The agent runs inside the customer’s own environment, including on‑premises and fully air‑gapped setups, connects to existing security tools without requiring schema normalization or proprietary pipelines, and autonomously investigates alerts, hunts threats, and documents every investigative step.

Why it matters: Security operations center teams can experiment with agentic automation on real incident queues without committing to a SaaS lock‑in or sending telemetry outside their own network perimeter. Because Crogl’s agent is designed to keep human analysts in control of decisions while handling repetitive investigation work, teams can test AI augmentation while preserving their existing workflows and governance.

Try/watch: Deploy the free agent in a lab or non‑production segment first, connect it to a subset of security tools, and compare investigation time, documentation quality, and false‑positive handling against your current manual procedures.