Amazon Bedrock AgentCore Adds Domain and Date Filters

Amazon’s Bedrock AgentCore just introduced runtime domain and publish-date filters for its Web Search feature. Developers can now restrict AI agent queries to specific websites or require results from recent sources. These filters operate at the level of each API request, giving organizations fine-grained control over the information their AI agents retrieve. That means a chatbot or workflow powered by AgentCore can reliably pull, say, only the latest information from an approved industry publication, rather than drawing from the noisy, sprawling entirety of the indexed Web.

This isn’t just a technical upgrade. For businesses, it addresses one of the perennial headaches of enterprise AI: trustworthiness of results. When an AI system can be told, on command, to ignore dubious blogs and stick to known-good sources—or only return news from the last week—the odds of embarrassing mistakes or stale recommendations drop.

Crucially, these new filters are enforced server-side. That’s a meaningful distinction for IT leaders. Client-side filtering, seen in some consumer-focused tools, can be skirted by crafty users or buggy implementations. Server-side filtering, by contrast, means the restrictions happen before the data ever lands in the AI workflow. For regulated industries, this helps support compliance requirements—think legal, finance, or healthcare, where data provenance matters.

In the projects we run, we've seen how server-side controls can be the difference between a demo and a production deployment. Enterprises want guarantees that their knowledge workers and automated agents draw from vetted pools, not just the open internet. AgentCore’s approach should give some comfort there, though outright perfection in source control is always aspirational in web-scale search.

Web Search for AgentCore is also now available in AWS’s Europe (Ireland) and Asia Pacific (Tokyo) regions. For customers concerned with data residency, latency, or regional regulatory requirements, this is table stakes. Expanding into these regions won’t redraw any competitive maps, but it fills necessary gaps for multinational organizations rolling out AI agents globally.

That said, regional support is only as valuable as the local content indexed. Many European and Asian businesses will want to test whether the search corpus actually reflects local language sources and up-to-date information—otherwise, the practical value may lag behind the headline.

Domain and publish-date filters sound like minor feature bumps. In practice, though, they target a real pain point for enterprise AI: the need for relevant, fresh, and trustworthy answers, not just plausible-sounding text. The payoff is about risk reduction as much as data quality. When a business can point its AI agent at a specific set of approved sites and demand recency, it lowers the chance of misinformation, hallucinations, or legal slips.

Still, the core challenge remains: no filter can perfectly guarantee that every pulled fact is accurate or contextually appropriate. Filters are only as good as the precision of the search backend and the curation of source lists. Companies should see these tools as safety rails, not absolute safeguards.

The new filters show AWS listening to enterprise concerns, but there are notable gaps. There’s no mention of more advanced controls—such as filtering by article author, structured source reputation scores, or content type. Nor is it clear how granular the published-date filters are (can you demand data from the last hour? The last five years?).

Competitors in the enterprise search space, including smaller upstarts, are starting to tout more sophisticated retrieval and filtering features. AWS’s moves are necessary, but not sufficient, for businesses that want truly explainable, auditable AI output. Expect further iterations—driven not by hype, but by customer demand for transparency and control.