Microsoft has introduced new security tools for IT admins and developers building AI products, including MDASH, an AI-powered scanning harness that integrates with Microsoft Defender. The company also released an integration between Microsoft Defender and GitHub Code Security, allowing developers to address risks early. Additionally, Microsoft introduced Agent 365 SDK, a tool for securing agents in various workflows and environments.
Microsoft has also unveiled its vision for an integrated agentic enterprise platform that combines various services for scalable AI deployment. The platform supports multiple AI models and emphasizes openness and flexibility, helping developers build, deploy, and govern AI agents safely and efficiently.
Other companies are also making strides in AI. OpenAI is expanding its robotics efforts, recruiting engineers for hardware and AI development to develop robots capable of performing useful tasks in the physical world. Meanwhile, Scale AI has partnered with companies to provide fine-tuning of AI models, making large-scale AI training and inference more feasible.
In the realm of AI applications, AdaCodec has been developed to reduce tokenization costs and latency in video encoding. This approach uses predictive visual coding to selectively transmit visual tokens, achieving superior performance at a fraction of the budget. Furthermore, AI is being used to help groups in conflict zones, not just to wage war, but to protect civilians, reunite families, and help broker ceasefires.
On the business front, a new report finds that AI is being shipped faster than customers can adopt it, highlighting the challenges of aligning marketing, sales, and education teams around a single hands-on experience. Companies like Amazon are being utilized for AI advancements, with Baz improving its AI Agent Code Review accuracy using Amazon Bedrock AgentCore.
Lastly, several new models and services have been released, including JetBrains' Mellum2, a 12B MoE model for fast, specialized tasks in multi-model AI pipelines, and Cybanetix's Managed AI Service, which provides a 360-degree view of the AI security ecosystem.
Key Takeaways
• Microsoft introduces new security tools, including MDASH and Agent 365 SDK, to help developers build secure AI products.• Microsoft unveils its vision for an integrated agentic enterprise platform for scalable AI deployment.
• OpenAI expands robotics efforts, recruiting engineers for hardware and AI development.
• Scale AI partners with companies to provide fine-tuning of AI models.
• AdaCodec reduces tokenization costs and latency in video encoding.
• AI is being used to help groups in conflict zones protect civilians and broker ceasefires.
• A new report finds AI is being shipped faster than customers can adopt it.
• Amazon Bedrock AgentCore is used by Baz to improve AI Agent Code Review accuracy.
• JetBrains releases Mellum2, a 12B MoE model for fast, specialized tasks in multi-model AI pipelines.
• Cybanetix launches a Managed AI Service to secure users, models, and agents in enterprise environments.
Microsoft Unveils New AI Security Tools
Microsoft has announced new security tools for IT admins and developers building AI products. These tools include MDASH, an AI-powered scanning harness that integrates with Microsoft Defender. The company also released an integration between Microsoft Defender and GitHub Code Security, allowing developers to address risks early. Additionally, Microsoft introduced Agent 365 SDK, a tool for securing agents in various workflows and environments.
Microsoft's Vision for Integrated AI Platform
Microsoft has unveiled its vision for an integrated agentic enterprise platform that combines various services for scalable AI deployment. The platform supports multiple AI models and emphasizes openness and flexibility. It is designed to help developers build, deploy, and govern AI agents safely and efficiently.
Cybanetix Launches Managed AI Service
Cybanetix has launched a Managed AI Service to secure users, models, and agents in enterprise environments. The service provides a 360-degree view of the AI security ecosystem, including observability, exposure mapping, behavioral monitoring, and runtime protection. It also offers AI risk assessment, technology deployment, and managed detection and response.
Baz Improves AI Agent Code Review with Amazon Bedrock
Baz has improved its AI Agent Code Review accuracy using Amazon Bedrock AgentCore. The company developed a multi-stage validation pipeline that combines code checking and dynamic runtime validation. This pipeline is powered by large language models served through Amazon Bedrock, providing scalable and secure AI inference.
Can Hackers Extract Fingerprints from Social Media Photos?
Experts say that hackers extracting fingerprints from social media photos is unlikely. While it is possible to pull fingerprints from photos, the risk is low for the average person. Hackers would need access to a physical scanner and would likely target high-value targets with access to secure facilities.
AdaCodec: Efficient Video MLLM Encoding
AdaCodec is a new approach to video encoding that reduces tokenization costs and latency. It uses predictive visual coding to selectively transmit visual tokens, achieving superior performance at a fraction of the budget. AdaCodec demonstrates marked improvements over baseline models in efficiency and performance.
OpenAI Expands Robotics Ambitions
OpenAI is expanding its robotics efforts and recruiting engineers for hardware and AI development. The company aims to develop robots capable of performing useful tasks in the physical world. OpenAI is seeking experts in hardware, operations, systems, and machine learning to support this initiative.
AI Helps Groups in Conflict Zones
Artificial intelligence is being used to help groups in conflict zones, not just to wage war. Organizations like the Red Cross and peace-focused tech companies are using AI to protect civilians, reunite families, and help broker ceasefires. AI is being used to analyze records, identify targets, and provide assistance.
AI Adoption Gap Widening
A new report finds that AI is being shipped faster than customers can adopt it. The report highlights the challenges of aligning marketing, sales, and education teams around a single hands-on experience. Organizations using hands-on labs are more likely to report developers reaching productivity within two months.
JetBrains Releases Mellum2 AI Model
JetBrains has released Mellum2, a 12B MoE model for fast, specialized tasks in multi-model AI pipelines. Mellum2 uses a Mixture-of-Experts architecture and is open-sourced under the Apache 2.0 license. The model is designed for code generation, editing, debugging, and conversational programming assistance.
Fine-Tuning AI Models with Modal
Benjamin Cowen from Modal discusses the shift towards custom, fine-tuned AI models and how serverless platforms simplify this process. Modal's infrastructure allows for unified GPUs and sandboxed environments, making large-scale AI training and inference feasible with less code and management overhead.
Sources
- Microsoft unveils new security tools for IT admins and developers building AI products
- Microsoft's Agentic Platform Vision
- Cybanetix unveils Managed AI Service to secure users, models, and agents
- How Baz improved its AI Agent Code Review accuracy using Amazon Bedrock AgentCore
- Can hackers pull your fingerprints from photos on social media? Experts explain.
- AdaCodec: Efficient Video MLLM Encoding
- OpenAI expands robotics ambitions, recruiting engineers for hardware and AI development
- How AI is helping groups in conflict zones where they’re seeking to bring peace
- AI Is Shipping Faster Than Customers Can Adopt It, New Research Finds
- JetBrains Releases Mellum2: A 12B MoE Model for Fast, Specialized Tasks in Multi-Model AI Pipelines
- Benjamin Cowen on Fine-Tuning AI Models with Modal
Comments
Please log in to post a comment.