Sunday, November 22, 2026 Edition No. 246
Search
Daily News Fully automated · 8 min read

AWS AI Updates: November 21, 2026

Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled

1. Web Search on Amazon Bedrock Adds Five Languages and a Country Field for Regional Results

Amazon Bedrock expanded Web Search, the capability that grounds agents and applications in cited, current web information, to Spanish, French, German, Italian, Japanese and non-US English on both Bedrock and Bedrock AgentCore. A new optional country field takes an ISO 3166-1 alpha-2 code such as “JP” or “DE” and biases results toward sources from that country, which helps with location-dependent questions like local regulations, prices or public officials. Existing customers get the language support without code changes, and pricing is the same in every language. Queries are processed in the Region that is called. Source

2. SageMaker JumpStart Adds a Model Leaderboard to Compare Foundation Models on Quality, Speed and Cost

Amazon SageMaker JumpStart now includes a model leaderboard in SageMaker Studio that compares foundation models side by side before deployment. AWS evaluates every model on BIG-Bench Hard, GPQA Diamond and MMLU-Pro with the same setup, and measures time to first token, throughput and instance cost for each JumpStart deployment configuration. Users can filter by provider, license and modality, and choose a use case profile and an optimization goal of balanced, cost, latency or throughput. The leaderboard launches with more than 70 large language models from providers including Meta, Google, Qwen, IBM and Arcee, at no additional charge. Source

3. Amazon Textract Insights Reaches General Availability for Splitting, Classifying and Extracting From Long Scanned Files

Amazon Textract announced general availability of Textract Insights, a generative AI feature that takes a long scanned file in one asynchronous request, splits it into sub-documents, classifies each one and extracts structured data. It targets insurance claims, tax packets and onboarding bundles. Users define fields in a JSON schema described in natural language or start from public blueprints such as W-2s, invoices and bank statements. Each value carries a confidence score and a page location, and the service can also return Markdown, HTML and plain text for RAG, summaries, figure captions and PII redaction. It requires no model training and is available in US East (N. Virginia) and US West (Oregon). Source

4. Amazon Textract Core Gets a Skill That Lets AI Agents Extract Document Data Without Custom Integrations

Amazon Textract published a skill for AI agents in Textract Core that lets an agent pull text, tables, forms and other data from documents from a plain-language request. The skill picks the right Textract Core API for each document, converts results into reading-order text and structured fields with a tested parser, and handles multi-page PDFs. High-confidence results are accepted and low-confidence ones are flagged for the user’s own review process. Agents using the AWS MCP server with the aws-core plugin can load it automatically, and it is also published in the AWS Agent Toolkit for agents such as Kiro, Claude Code and Cursor. There is no additional charge beyond standard Textract pricing, and it is available in 16 commercial Regions. Source

5. AWS Quick Adds Enterprise Guardrails in Preview for Content Safety and Sensitive Data in Prompts

AWS Quick introduced enterprise guardrails in preview, giving administrators controls over content safety and sensitive information in user prompts. Policies can be assigned to a user, an identity provider group or the account as a default, with user policies overriding group policies and group policies overriding the default. Administrators build them in the Quick console from content filters, denied topics, blocked words and phrases, prompt-attack detection and PII controls. A detect-only mode logs matches without blocking so rules can be tested on live traffic, and enforcement then blocks matching prompts and shows the user a message. Events can be reviewed in Amazon CloudWatch Logs, and the preview is open to select Professional and Enterprise customers. Source

6. Bedrock AgentCore Runtime Instances Add Amazon EC2 G7 GPU Instances

Amazon Bedrock AgentCore runtime instances now support Amazon EC2 G7 instances, which use NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, for agents that do 3D rendering or model inference. G7 sizes range from g7.2xlarge to g7.48xlarge and are enabled by adding them to the allowed instance types of a capacity provider. AgentCore installs the NVIDIA drivers, so standard container images can run GPU workloads without bundling them. It is available in eight Regions including US East (N. Virginia), US West (Oregon), Europe (Frankfurt) and Asia Pacific (Sydney), with management charges on top of standard EC2 costs. Source

7. AWS Agent Registry Lets Agents and Tools Carry Verified Domain Names and Be Found Publicly

AWS Agent Registry now supports public naming, which attaches verified domain names to agents and AI tools. Developers and network administrators complete domain verification to prove ownership, and published records are tied to that domain identity so others can confirm an agent’s source before connecting. Public records can be searched by any AWS-authenticated user with semantic and lexical search. There is no additional charge at this time, in all Regions where Agent Registry is available. Source

8. AWS Agent Registry Adds WebBotAuth Records So WAF Bot Control Can Tell Verified Bots From Malicious Traffic

AWS Agent Registry can now register WebBotAuth bots as a new record type. WebBotAuth is an IETF standard for verifying bots, and a registered bot proves domain ownership through it. The verification works with AWS WAF Bot Control, so rules can combine a bot’s verification status with its identity, for example allowing WBA-verified search crawlers while blocking unverified traffic. It is aimed at content providers and application developers and is available in all Regions where Agent Registry is available. Source

9. AWS DevOps Agent Expands Chat to MCP Clients, Teams and a Compute Sandbox

AWS DevOps Agent expanded its chat capabilities. The agent now works across MCP clients such as Kiro and Claude Code as well as Slack and Microsoft Teams, and users can direct it through MCP tools to act in GitHub, GitLab, Azure DevOps, ServiceNow and PagerDuty. Teams gets two-way conversation, the agent creates a dedicated Slack code channel per incident, and a sandbox lets it run scripts and reach the network. Chats can be renamed and archived on every supported platform. Source

10. AWS DevOps Agent Gets a Curated Open Source Repository of Skills, MCP Servers and Partner Connectors

AWS DevOps Agent now has an AWS-curated open source tools repository of skills, MCP servers, connectors and custom agents that users can import to improve investigations and add operational context. Contributions come from the community, partners and AWS service and support teams. Partner connectors listed include Bluebox, Bronto, Celonis, Clickhouse Observability, Coralogix, Cribl, Elastic, Halo, Honeycomb, LaunchDarkly, Neo4j, Sumo Logic and Wiz. Users browse the catalogs from within their DevOps Agent instance and can share their own tools through a pull request. Source

11. Amazon Transcribe Adds TranscribeAudio, a Synchronous API That Returns a Full Transcript in One Request

Amazon Transcribe added a synchronous transcription API, TranscribeAudio, that takes a complete audio file in one request and returns the full transcript in the response, with no job to start or poll. AWS says accuracy matches batch transcription. It supports word-level timestamps, language identification, speaker diarization, two-channel identification, custom vocabulary and PII redaction, for audio from 0.5 seconds to 1 hour, up to 25 MB inline or 500 MB from Amazon S3. Pricing is $0.0070 per minute of audio billed per second, with PII redaction as an add-on, and it is available in 14 Regions. Source

12. Amazon DynamoDB Vector Search Filters Now Support Ranges, IN Lists, Existence Checks and OR/NOT

Amazon DynamoDB extended inline filters in vector search beyond equality. The SearchVectors search condition expression now accepts comparison operators and BETWEEN on Number attributes, the IN operator, the attribute_exists and attribute_not_exists functions, and AND, OR and NOT. DynamoDB applies the filter during the search, so every result returned matches. Example uses are finding products within a price range and set of brands, and limiting RAG results to document categories published after a given date. It is available in all Regions where DynamoDB vector search is offered. Source

13. Amazon S3 Vectors Raises Filterable Metadata to 40 KB per Vector, Up From 2 KB

Amazon S3 Vectors now supports up to 40 KB of filterable metadata per vector, up from 2 KB, so RAG, agentic and semantic-search applications can filter on more attributes such as category, permissions, timestamps and tags. Indexes in newly created vector buckets get the higher limit by default, and existing indexes can be switched to enhanced index mode in place with the UpdateIndexMode API. PutVectors and QueryVectors work unchanged, and there is no additional cost in any Region where S3 Vectors is offered. Source