AWS Launches Two New Services to Fix AI Agent Blind Spots
Amazon Web Services used its Summit in New York to address one of the most persistent problems with enterprise AI deployment: agents that move fast but make costly mistakes. The company launched two new services targeting AI agent business context and security, two areas where current AI tools routinely fall short.
The first, AWS Continuum, automates the full lifecycle of code vulnerability management, from detection and prioritization to validation and remediation. AWS noted in its security blog that specialized AI models can identify vulnerabilities and map attack paths faster than human defenders can respond, making traditional security workflows inadequate. Continuum starts in a learning mode that requires human approval before any fix is applied, then can graduate to an enforcement mode where it resolves defined issues automatically.
The second service, AWS Context, builds a shared knowledge graph from existing enterprise data including databases, documents, emails, and chat messages and makes it available to every agent across an organization. Without this layer, AWS argues, agents operating without AI agent business context too often produce confident but incorrect recommendations. The service derives relationships between data points, applies business rules, and enforces access controls so agents only retrieve information they are authorized to use. It is built on the same knowledge graph foundation as Amazon’s AI assistant Quick.
AWS also announced upgrades to its DevOps Agent, which is gaining a Release Readiness Review feature that checks every code change against production requirements before deployment. A second feature generates a test plan from each code change and runs it in a production-like environment. The preview is available free in the US East region. These additions follow a period of internal turbulence at Amazon, where AI coding tools were reportedly linked to at least two AWS outages, including a 13-hour incident caused by Kiro deleting and rebuilding an environment. The company subsequently required senior engineers to approve all AI-generated code.
On the product side, AWS is bringing its coding agent Kiro to iOS, giving users a mobile interface to monitor tasks, review code changes, and grant approvals while sessions continue running in the cloud. Bedrock AgentCore, the company’s platform for running agents in production, is also receiving a managed knowledge base with connectors to S3, SharePoint, Confluence, and Google Drive, alongside built-in web search and security filters capable of detecting manipulative prompts, malicious content, and data leaks. Integrations with third-party security providers including Check Point, Zscaler, and SentinelOne are also planned.
