AWS adds built-in Web Search to Amazon Bedrock for foundation-model grounding

AWS made Web Search generally available on Amazon Bedrock, a server-side tool that lets foundation models ground their responses in current web knowledge without integrating a third-party search provider. Developers enable it by setting a single parameter in an existing OpenAI-compatible API call, and responses come back with structured citations — a low-friction path to reducing hallucination and stale-knowledge problems for models whose training cutoffs lag real-world events.

The capability was previously available only within Bedrock AgentCore; this release brings native grounding to the broader Bedrock foundation-model surface. That matters because grounding is table stakes for enterprise deployments where answers must be current and citable — legal, financial, and support workflows in particular. By offering it first-party, AWS removes the need to wire up external search APIs and manage their keys, rate limits, and data-handling terms.

Competitively, built-in grounded search puts Bedrock in line with capabilities OpenAI, Google, and Perplexity already offer, and complements AWS's broader week of agent infrastructure announcements — runtime instances, Dogwood governance, and the n8n integration harness. It also pairs naturally with AWS's aggressive Bedrock pricing moves, including an 80% cut to hosted GPT-5.6 Luna input pricing.

The practical questions for developers are which search backend powers the tool, how citation quality and coverage compare to dedicated providers like Tavily or Exa, and what per-query pricing looks like at scale. As a single-parameter add to an existing call, adoption friction is minimal — the interesting signal will be whether teams standardize on Bedrock's native grounding or keep specialized search providers for higher-quality retrieval. For AWS, the strategic logic is keeping the entire RAG-and-grounding loop inside its walls.