During re:Invent 2025, AWS announced its largest AI model expansion to date for its generative AI service platform, Amazon Bedrock, adding 18 new open weight models. In addition, AWS also upgraded Amazon GuardDuty and AWS Security Hub to enhance threat detection and risk correlation analysis capabilities through AI technology, thus improving cloud security.
Amazon Bedrock incorporates Mistral Large 3, Gemma 3, and even includes OpenAI tools and the newly built Amazon Nova 2 series of models.
This update to Amazon Bedrock reinforces its position as a "model marketplace," allowing businesses to flexibly select and switch between the most suitable AI models without rewriting code. The lineup of 18 newly added AI models is as follows:
• Mistral AI:The first release is Mistral Large 3, Mistral AI’s most advanced open weighted model, optimized for long text, multimodal, and instruction reliability. It also includes the Ministral 3 series, which is designed for compact general AI applications and offers versions with 30 billion, 80 billion, and 140 billion parameters respectively.
• Mainstream and popular AI models:Added features including Google Gemma 3, MiniMax M2, and NVIDIA Nemotron.
• OpenAI:Joining OpenAIOpen weighted language model gpt-oss Safeguard.
• Amazon Nova 2:Joining at the same timeAWS's own Nova 2 series modelsThe series includes Lite, Pro, and Sonic versions, emphasizing the cost-effectiveness of industry-leading models in inference, multimodal processing, and conversational AI. In addition, Amazon Nova 2 Omni, which can perform multimodal inference and image generation with a single model, is launched and is available for preview use through the AWS service platform starting today.
To enable businesses to use these models more flexibly, AWS also launched a pioneering "open training" customization feature called Nova Forge, which allows organizations to access pre-trained model checkpoints and fine-tune them with proprietary data.
Meanwhile, Nova Act achieves a groundbreaking 90% reliability in automated workflows within the browser UI, and has already been adopted by customers such as Reddit and Hertz.
AgentCore Upgrade: Giving AI Agents "Memory" and "Strategy"
In building AI agents, Amazon Bedrock AgentCore has released several preview features to help developers create agent services that are closer to production-ready:
• Policy:It allows teams to use natural language to set clear boundaries for the actions of agents.
• Evaluations:It provides 13 pre-built evaluators to continuously monitor the correctness and security of agent behavior.
• Memory:New plot features are introduced, allowing the agent service to learn from past experiences and become increasingly intelligent.
Partnering with Adobe, Sony, and WRITER to deepen applications in vertical industries
During this demonstration, AWS also showcased how its infrastructure enables various businesses:
• Adobe: The two companies announced a restructuring of their creative and marketing processes, utilizing Adobe Firefly's text-to-image and text-to-video models, which are currently being trained on AWS EC2 P5 and P6 instances. Additionally, Adobe is exploring the use of Amazon Bedrock AgentCore to accelerate autonomous agent execution capabilities within its products.
• Sony: It has built two major platforms using AWS, including an internal enterprise AI platform (powered by Amazon Bedrock AgentCore) that serves 5.7 Sony employees, and the Sony Engagement Platform that connects electronic products, PlayStation game consoles and audio-visual content. The latter further deepens online fan connections and interactions by processing data from more than 500 sources through Sony Data Ocean, which is built on the AWS platform.
• WRITER: This enterprise AI technology provider integrates with the Amazon Bedrock platform, enabling its customers to directly access models included in Amazon Bedrock within the WRITER platform. WRITER has also launched a new "agent supervision suite" as a control center for enterprise AI, providing detailed monitoring and centralized auditing workflows to address the trust issues that enterprises are most concerned about when expanding AI agents.
GuardDuty Extended Threat Detection Adds Support for Amazon EC2 and ECS
In terms of cybersecurity, the Amazon GuardDuty Extended Threat Detection feature has now been officially expanded to support Amazon EC2 and Amazon ECS environments.
Previously, this feature was primarily offered for IAM, Amazon S3, and EKS, but it now extends to virtual machine and container workloads. GuardDuty leverages AI and machine learning models trained at AWS scale to correlate scattered signals such as anomalous program creation, persistent intrusion attempts, or mining activities into a single high-severity event, and maps them to the MITRE ATT&CK framework, helping cybersecurity teams accelerate their response to attack incidents.
The full version of Security Hub provides near real-time threat correlation.
In addition, new features for AWS Security Hub have been announced to be available in general. The service now offers near real-time risk analysis, automatically linking signals from various tools such as GuardDuty, Amazon Inspector, and Macie.
Through enhanced visualizations and historical insights, Security Hub helps enterprises identify potential attack paths, understand how threats, vulnerabilities, and misconfigurations chain together to create risks, and quickly prioritize their handling, thus addressing the pain points of managing complex multi-security tools.
