AI news, product launches and company developments.
OpenAI & AnthropicBaz improved its AI agent for code review accuracy by using Amazon Bedrock and AgentCore. The blog details their architecture and implementation process. This advancement helps automate and enhance code review workflows effectively.
Amazon Bedrock AgentCore Identity now supports referencing secrets stored in AWS Secrets Manager. This allows organizations to use their existing secrets management processes with AgentCore. It also supports secrets from other AWS accounts and third-party secret managers via external connectors.
OpenAI's GPT-5.5, GPT-5.4, and Codex models are now generally available on Amazon Bedrock. These models can be deployed in production applications using Bedrock's high-performance inference engine. This availability enables easier integration of advanced AI capabilities into enterprise solutions.
A recent study reveals that major AI models from OpenAI, Anthropic, Google, Amazon, and xAI fail against a specific type of attack. Current safety benchmarks used by enterprise buyers do not effectively measure this vulnerability. This highlights a critical gap in AI model evaluation methods.
Amazon Bedrock AgentCore Gateway now supports extended Model Context Protocol (MCP) features for better access control and security. It centralizes credential management and observability for enterprises deploying MCP servers. This enhancement helps organizations securely manage AI model usage at scale.
Amazon Bedrock AgentCore gateway now supports securing AI agents using Policy for deterministic access control and Lambda interceptors for dynamic validation. The blog demonstrates combining these features to enforce geography-based access control. This enhances security and control over AI agent operations in data environments.
Amazon Bedrock AgentCore Payments introduces built-in guardrails to ensure safe agentic payment systems. The solution addresses key risks in designing autonomous payment workflows. This advancement helps secure financial transactions managed by AI agents.
Amazon Bedrock AgentCore introduces AgentOps to help manage agentic AI at scale. Agentic AI systems make autonomous decisions, causing unique operational challenges like unpredictable costs and debugging difficulties. AgentOps adapts DevOps practices to deploy, manage, and improve these AI agents effectively in production.
AWS introduces GPUDirect support on Amazon FSx for Lustre combined with TurboQuant to speed up loading large language models into GPU memory. This reduces wait times for GPUs to be ready for inference, especially for models with hundreds of billions of parameters. Faster model loading enables more efficient iteration and deployment of LLMs on AWS GPU...
OpenAI's frontier models and Codex are now available on AWS. This integration allows enterprises to use OpenAI's tools within AWS environments and workflows. It simplifies adoption and accelerates moving from evaluation to production.
Amazon SageMaker now offers comprehensive observability for AI LLM inference using Amazon Managed Grafana dashboards. This solution monitors GPU utilization and LLM quality metrics in real time. It helps users optimize performance and maintain high-quality outputs for deployed models.
Azercell Telecom collaborated with AWS to train Azerbaijani large language models using Amazon SageMaker AI. They addressed challenges related to the language's complexity and limited data. This effort resulted in a production-ready framework for telecom applications and customer chatbots.
