AWS launched TorchNeuron at the 2026 PyTorch North America conference. The native PyTorch backend is designed to run existing code on AWS Trainium AI chips with little modification, according to InfoQ. Its documentation says it natively supports distributed training and parallel APIs including DDP, FSDP2 and DTensor. AWS says a configuration change is sufficient to switch hardware, without rewriting or revalidating models. The report also says TorchNeuron can share hardware capacity between training and inference workloads and reallocate resources from inference to training during off-peak periods.
AWS Machine Learning Blog reports that AWS Quick now offers a Meta Ads AI connector built on ads Model Context Protocol server operations. Marketers can use natural language to analyze campaigns, create or edit ads, build product catalogs, and combine Meta data with other sources for dashboards and automated workflows. Quick asks for permission before making changes such as creating campaigns or adjusting budgets.
The Financial Times reported that Amazon is nearing a $7.5 billion deal to buy AI startup Decart. Talks are at an advanced stage, and an agreement could come together as soon as next week. The Financial Times said the acquisition would help AWS compete with Nvidia in the race for computing power.
AWS released updates to its AWS Quick AI workspace covering conversational workflow creation, deterministic and longer-running flows, concurrent multi-agent orchestration, and continued background execution after a browser tab closes. The web experience also gains desktop agents and skills, reusable skills, additional data connectors, more languages, and administrative controls for prompt safety, permissions, and data sources. Most capability details are vendor claims, while AWS said the features are available in all regions where AWS Quick is available.
AWS released a distributed KV caching skill for the Agent Toolkit for AWS that helps coding agents evaluate, configure, and troubleshoot caching for self-hosted LLM inference. The skill works with Kiro, Claude Code, OpenAI Codex, and other agents supporting Agent Skills, and can use KVTrace to estimate workload-specific savings before deployment. AWS says Amazon ElastiCache can reduce inference costs by up to 40% and time to first token by up to 80%, but these are workload-dependent vendor claims rather than universal guarantees.
AWS introduced the SageMaker HyperPod Global Inference Gateway, a managed endpoint that routes inference requests across registered clusters within one Region or across multiple Regions. It selects clusters using KV cache headroom, queue depth, Region proximity, and error rate, while providing automatic failover, SigV4 and IAM authentication, and PrivateLink connectivity. AWS says the gateway is available in 13 Regions, but its performance and reliability benefits remain vendor-reported.
AWS announced the general availability of Amazon Textract Insights, a new generative AI API for extracting structured fields from documents and mixed document packets. Users can define fields in natural language and receive values with confidence scores and page-level grounding without training a model, according to the AWS Machine Learning Blog. The service supports PDF, PNG, JPEG, and TIFF inputs, processes documents in six languages, and is priced per page.
OpenComputer was introduced as an open-source cloud coding agent that teams can run in their own AWS, GCP, or Azure accounts, either through its free cloud service or by self-hosting. Its founder says the agent harness and sandbox share a durable VM, sessions use Git worktrees, and session files are backed up after every turn for resumption on another VM.
Baseten says it secures more than 100 inference clusters across more than 20 cloud providers, including hyperscalers and neoclouds. Its security approach includes per-cluster hardening, network and tenant isolation, scoped credentials, isolated build environments, runtime detection, centralized logging, and an accompanying checklist for evaluating security standards.
Standard Bank Group, working with AWS Professional Services, deployed an automated assessment that scores generative AI use cases against international standards and internal policies. AWS Machine Learning Blog reports that the system uses an unmodified foundation model on Amazon Bedrock as an LLM-as-a-judge to collect evidence, identify mitigations, produce reports and a dashboard, and support existing risk-committee approval gates.
AWS Machine Learning Blog reports that Kimi Code CLI v2.7.0 adds a native Amazon Bedrock provider that uses the standard AWS credential chain, including named profiles and SSO. The integration can connect Kimi K3.1 Pro Highspeed on Amazon Bedrock without separately managing a Bedrock API key, although suitable AWS credentials, IAM permissions, region, and model access are still required.
Researchers at Stony Brook University reported that automated scanners searched public container images and extracted embedded cloud credentials. In a four-month honeypot experiment spanning eight registries and 300 images, the team recorded more than 75,000 image pulls and over 140,000 AWS test-key API calls. Some OpenAI and Anthropic test keys were also used against AI model APIs. The fastest AWS test-key call occurred 7.86 seconds after publication. Based on key-use records and container execution status, the researchers inferred that static analysis allowed the scanners to retrieve keys without running the containers.
AWS Machine Learning Blog describes how BMW Group’s R&D organization is building, governing, and operating hundreds of agents used by thousands of developers. The reported approach combines a central agent registry, framework-agnostic integration-layer controls, specialized agents, reusable infrastructure-as-code blueprints, and automated security and compliance checks. AWS and BMW say these practices speed deployment, reduce duplicate work, and maintain quality, although those benefits are not independently verified in the report.
AWS says Amazon Connect Talent uses structured questions, adaptive follow-ups, and evidence-based competency scoring to conduct job-related interviews at scale. The service separates evidence gathering from evaluation, assigns scores on a fixed 1–5 formula, and flags competencies when transcripts lack sufficient evidence. In a vendor-reported review by an independent industrial-organizational psychology firm, its scoring agreed with expert decisions in 89.3% of 252 rubric signals across 36 interviews, with a Cohen’s kappa of 0.79.
Amazon Polly voice cloning is now generally available as a gated release, creating a custom text-to-speech voice from a few minutes of recordings from a speaker who has consented to being cloned. It supports English, Spanish, German, French, Italian, Portuguese and Hindi, with each cloned voice speaking the language in which it was recorded. Amazon Polly says the voices run on its generative engine with similar naturalness and accuracy to its other generative voices. Voice data is encrypted and access-controlled. Access has been enabled for selected customers, while others must contact their AWS account team or apply.
AWS announced that cross-Region inference routing is generally available in Amazon SageMaker AI for real-time and asynchronous inference in most commercial AWS Regions. Its inference router provides one invocation path across registered endpoints and inference components in different Regions and accounts, preferring healthy local capacity before shifting new requests and gradually restoring local traffic after recovery. AWS says the routing feature adds no fee beyond standard SageMaker AI and CloudWatch charges, while response streaming is unsupported and some quotas and regional exclusions apply.
AWS Machine Learning Blog announced that PolyAI Dialog-RSN-1, a reasoning model for real-time complex conversations, is available through AWS Marketplace and Amazon SageMaker JumpStart. Customers can subscribe to the model and deploy it as a SageMaker endpoint within their AWS account or through SageMaker Studio. PolyAI claims the model directly processes caller audio, reasons about caller needs, and responds in real time, though these performance claims are vendor-sourced.
36Kr reports that some developers are renting physical Mac mini systems in the United States or Australia to access Claude Code, while others are moving Codex or Claude Code environments to overseas cloud hosts for unattended automation. The article attributes these workarounds to tighter Claude enforcement in unsupported regions and anecdotes of accounts being disabled without refunds, but provides no Anthropic statement or data establishing their prevalence. It also discusses Google Cloud credits and recommends commercial APIs, adequately resourced virtual machines, and remote development workflows.
Elastic Security Labs compared five open-source AI agentic attack frameworks—ARTEX, CyberStrikeAI, Cairn, PentAGI and HexStrike—following their reported misuse in real intrusions. The research says they can plan attack steps, invoke security tools, delegate work among agents and install required tools, while citing observations of Cairn in actual intrusions, a PentAGI-based reconnaissance platform and a CyberStrikeAI server found during an AWS investigation. Elastic recommends behavior-based detection because rapid project updates can quickly invalidate indicator-based rules.
Barracuda reported that attackers used the AI penetration-testing framework ARTEX to move across 10 cloud accounts over 44 hours, harvesting credentials, exploring internal systems, probing data-analysis and source-code systems, and targeting Kubernetes clusters. The initial access came from a cloud key issued to a third-party AI company and not revoked after the project; attackers also pushed short-lived branches to expose build secrets, while Barracuda found no direct evidence of XDR customer-data access but said incomplete records prevented it from ruling that out.
Malicious versions 1.26.1 and 1.26.2 of Microsoft’s Agent Framework Python package were uploaded to PyPI on November 17, according to Wiz and other security researchers cited in the reports. The malware was designed to run when Python started, even without importing the package, and collect credentials, cloud tokens, local configuration files and information about running processes. It targeted services including Azure, AWS, GitHub, OpenAI, Docker and Hugging Face. Version 1.26.2 also sought GitHub Actions OIDC tokens, while both releases were removed from PyPI. Researchers said it remains unknown whether the attack succeeded or whether any developers lost secrets. They advised users who installed either version to remove it, revoke and replace exposed secrets, and investigate cloud accounts and automated build environments, including GitHub Actions.
Amazon SageMaker AI Serverless Model Customization now supports Full Fine-Tuning (FFT), allowing customers to update all model parameters without provisioning GPUs, configuring clusters, or setting up distributed training. AWS Machine Learning Blog says FFT can outperform LoRA for complex tasks with sufficiently large, high-quality datasets, but its own tests found it was highly sensitive to hyperparameters and not universally superior. LoRA scored 81.5% on LLM-as-Judge versus 79.5% for FFT, while aggressive FFT learning rates sharply degraded MATH and MMLU results.
AWS announced the Amazon SageMaker JumpStart Model Leaderboard, a new SageMaker Studio view for comparing JumpStart models side by side on quality, speed, and cost. According to the AWS Machine Learning Blog, it launches with more than 70 models from providers including Meta, Google, Qwen, IBM, and Arcee, using AWS-run benchmarks and standardized SageMaker deployment tests. Users can filter models by license, provider, and input and output modality, then test shortlisted models with their own data and prompts.
AWS Machine Learning Blog reports that thousands of engineers and engineering-adjacent staff at Airbnb use the internal AirChat tool to run coding agents each week. The agents send requests through Airbnb's LLM gateway on Amazon Bedrock, where the company evaluates models using its own tasks and cost per task before rollout, then monitors production latency, throttling, token consumption, and spend. Airbnb also uses evaluation results to identify equivalent models and rebalance heavy users when costs rise.
AWS Machine Learning Blog says Amazon SageMaker AI cross-Region inference routing is generally available as of November 18, 2026, in commercial Regions supporting real-time endpoint inference. Salesforce uses the feature to route requests across its existing endpoints in multiple Regions and accounts, adding serving capacity while preserving data residency commitments. The router does not create capacity or copy models, and operators can cap remote traffic to protect capacity reserved for local requests.
AWS Machine Learning Blog says Social Department built QuickSearch, a production system that retrieves entertainment footage for shareable scenes, quotable quotes, memeable moments, and trend alignment. Its AWS architecture combines TwelveLabs Marengo 3.0 and Pegasus 1.2 models on Amazon Bedrock, Claude models for interpretation and editorial ranking, and hybrid retrieval through Amazon OpenSearch Serverless. The system uses source-grounding checks, Amazon Bedrock Guardrails, and human approval before publishing.
AWS Machine Learning Blog published a tested architecture pattern for connecting Amazon Bedrock AgentCore Gateway to private APIs and MCP servers through a hub-and-spoke network with centralized firewall inspection. The pattern places an Amazon VPC Lattice resource gateway in the network account and includes six configuration decisions, validation checks, CloudFormation templates, test scripts, troubleshooting guidance, quotas, and an address plan. AWS built and tested it in us-east-1 using three member accounts and an AWS Organizations management account, and recommends adapting it to each organization’s account architecture.
AWS held its inaugural AWS Quick User Group DC meetup at the AWS Skills Center in Arlington, Virginia, where participants explored using governed enterprise data for AI analytics and agentic workflows. Financial Independence Group described building toward AI-powered analytics with AWS Quick, while attendees created no-code apps and six participants built multi-step, event-triggered AWS Quick Flows.
At an investor briefing, Auras Technology chairman Lin Yu-shen said new ASIC customers including AWS and Meta would be the company’s main growth drivers next year. The company expects server products to rise from 80% to 90% of revenue and ASIC revenue to exceed revenue from the Nvidia-led GPU business. Management also forecast that 2027 revenue growth would be no lower than this year’s estimated growth of more than 70%, with margins improving. Equipment spending for expansion will be reviewed quarterly against customer demand; these financial figures are management forecasts.
A Juejin analysis compares CAIE Artificial Intelligence Engineer Level I with AWS Certified Machine Learning Engineer – Associate. It positions the former around general AI use for content processing, data analysis, and business collaboration, while the latter covers machine learning engineering in AWS environments, including data preparation, deployment, monitoring, and troubleshooting. The article also warns that “application-oriented” and “technical” are practical classifications rather than official certification levels.
What problem it solves
Developers using AI coding CLIs need a way to delegate work among multiple agents and coordinate their execution. CAO addresses this by running provider CLIs in isolated terminal sessions and giving a supervisor tools for coordinating workers.
Key features
Multi-agent delegation — A supervisor delegates work to specialist agents, with parallel and sequential execution suppor…
What problem it solves
AI-generated code can violate the rules its developers intended, especially when those rules are described in natural language and the resulting code is not reviewed closely. Bend addresses this by representing important rules as laws and requiring mathematical proofs when the code is edited.
Key features
Proof-enforced application laws — Applications can declare rules in `L…
What problem it solves
Agents can understand diagram content but often produce generic layouts that do not match a site's visual identity. Diagram Design addresses this by applying a defined editorial style to agent-generated diagrams and producing publishable, self-contained files without a manual design workflow.
Key features
Self-contained HTML and SVG output — Each diagram is written as one HT…
What problem it solves
Organizations need a distributed storage layer that can serve objects, files, tables, and vectors while keeping administration and horizontal expansion straightforward. SeaweedFS addresses this by providing these capabilities on shared storage and allowing capacity to grow by adding volume servers.
Key features
Unified distributed storage — A single cluster exposes S3 object…
What problem it solves
Modern systems need controlled access to database credentials, API keys, and other secrets across platform-specific infrastructure. Teams also need secure storage, key rotation, encryption, revocation, and detailed audit logs without building a custom solution.
Key features
Secure secret storage — Vault stores arbitrary key/value pairs and encrypts data before writing it to …
What problem it solves
VPN traffic can expose protocol characteristics that are examined during network inspection. The project addresses this by encrypting UDP tunnels, applying configurable traffic masks, and rotating session keys while providing bounded packet handling and live traffic-quality checks.
Key features
Encrypted and forward-secret sessions — The InitialV2 exchange confirms a fresh s…
What problem it solves
Leaked credentials can be difficult to identify and assess because they may be stored across repositories, chats, wikis, logs, filesystems, and other systems. Finding a credential pattern alone does not establish whether it is active, so the project combines discovery, classification, validation, and analysis.
Key features
Broad secret discovery — Scan sources including Git …
What problem it solves
Creating intelligent AI agents that can collaborate on complex tasks can require code and substantial integration work. Docker Agent addresses this by letting users define agents in YAML, assign tools, and run them through the Docker CLI.
Key features
Multi-agent orchestration — Create teams of specialized agents that delegate tasks automatically.
Source: README
Extensible t…
What problem it solves
AI coding agents need AWS-specific tools, knowledge, and guardrails to work effectively with AWS services. Teams using different coding agents also need a consistent way to configure AWS API access, guidance, and reusable skills.
Key features
AWS MCP Server — The managed MCP Server gives agents access to AWS APIs, isolated script execution, and current documentation through …
What problem it solves
Organizations need to build and maintain custom applications such as dashboards, admin panels, customer 360 tools, IT automations, and service management systems. This platform streamlines those application-development workflows while supporting databases and APIs.
Key features
Low-code custom application development — Build internal applications through a low-code platform …
What problem it solves
Ad-hoc AI coding can lose context as projects grow. AI-DLC keeps requirements, decisions, implementation, tests, and operational work connected through an audited lifecycle.
Key features
Harness-neutral workflow core — A single core runs across multiple coding-assistant harnesses, with thin harness-specific integrations.
Source: README
Structured delivery lifecycle — The wor…
What problem it solves
Teams need to capture, search, and analyze multiple observability signals without managing fragmented tools or complex infrastructure. OpenObserve addresses this with a unified platform, a single-binary deployment option, and familiar SQL and PromQL query languages.
Key features
Unified observability — Logs, metrics, traces, RUM, dashboards, alerts, incidents, pipelines, and…
What problem it solves
Developing against AWS services often requires a cloud account, authentication token, or paid feature gates. Floci provides AWS-shaped services locally so existing AWS SDKs, CLI commands, infrastructure-as-code workflows, and test suites can run without those dependencies.
Key features
Local AWS services without an account — Run AWS-compatible services locally without a clou…