FBI Doubles Down on AI for Investigations and Operations

The Federal Bureau of Investigation (FBI) is significantly expanding its use of artificial intelligence in key areas, including criminal investigations, intelligence analysis, and internal operations. According to the Department of Justice’s (DOJ) latest public inventory, the number of AI use cases reported by the FBI has more than doubled to 50. These AI applications encompass machine learning, computer vision, and generative AI, primarily aimed at labor-saving tasks like translation, transcription, and summarization. The 2025 DOJ AI Use Case Inventory reveals a sharp increase in the FBI’s AI tools for law enforcement, jumping from 15 to 27 in just one year, reflecting a broader 30.7% increase in AI adoption across the entire DOJ.

Sources:

Duke Engineers Create ‘Lego-Like’ Programmable Material for Adaptive Robotics

Mechanical engineers at Duke University have developed a novel programmable material with the potential to revolutionize robotics. The system, which functions like advanced Lego blocks, can alter its mechanical properties on demand. It consists of small cells filled with a gallium-iron composite that can be switched from solid to liquid using an electrical current. This allows researchers to “write” specific mechanical properties into the material by liquefying cells in any desired pattern. To showcase the technology, the team constructed a robotic fish with a tail made from these programmable cells. By changing the pattern of solid and liquid cells, they successfully altered the fish’s swimming trajectory without changing the motor’s input. This innovation paves the way for robots that can adapt their stiffness and flexibility for diverse tasks and environments, with potential applications in adaptive medical devices.

Collins Aerospace’s Sidekick AI Successfully Flies Uncrewed Jet in CCA Program Test

Collins Aerospace, an RTX business, has successfully tested its Sidekick mission autonomy software on a YFQ-42A uncrewed jet from General Atomics Aeronautical Systems, Inc. The test, conducted as part of the U.S. Air Force’s Collaborative Combat Aircraft (CCA) program, featured a four-hour autonomous flight overseen by a ground operator. The flight validated the seamless integration of the Sidekick software with the aircraft’s mission systems, enabling it to execute precise piloting commands. The Sidekick solution is engineered to enhance open systems collaboration between human-led teams and autonomous platforms in complex combat air operations.

Sources:

Google Releases Gemini 3.1 Pro, Boosting Advanced Reasoning for Developers

Google has officially launched its next-generation flagship model, Gemini 3.1 Pro, featuring significantly enhanced capabilities for advanced reasoning and complex task management. The new model is now rolling out in preview to developers and enterprise customers via the Gemini API, AI Studio, Vertex AI, and Android Studio. Subscribers to Google AI Pro and Ultra plans can also access Gemini 3.1 Pro through the Gemini app and NotebookLM. Google reports that the model shows dramatic performance improvements on complex problem-solving benchmarks, doubling the reasoning performance of its predecessor in certain tests. The current preview phase is designed to gather user feedback and validate updates before a wider general release.

Sources:

Researchers Propose ‘Prompt-to-Drug’ AI Framework for Autonomous Drug Discovery

A new perspective published in ACS Central Science by researchers from Insilico Medicine and Lilly outlines a framework for a fully autonomous, AI-driven drug discovery process. This “prompt-to-drug” concept imagines a central AI controller capable of managing the entire R&D pipeline—from target discovery and molecular design to automated synthesis and clinical trial planning—all initiated by a simple prompt from a scientist. The proposed system would integrate generative AI, multimodal foundation models, and automated lab systems to dramatically accelerate pharmaceutical research. The authors suggest that while the core components for this vision exist, achieving a true end-to-end autonomous system will require extensive collaboration between academia, biotech firms, and regulatory bodies.

Sources:

California’s ‘Digital Dignity Act’ Leads New Wave of US AI Legislation

Legislative activity around artificial intelligence is intensifying across the United States, with California introducing the Digital Dignity Act (SB 1142). This bill aims to combat the use of digital replicas for false impersonation by requiring platforms to offer a way for users to revoke access to their digital likeness created by generative AI. Other states are also advancing AI-related laws: an AI safety bill in Oregon (SB 1546) has passed the Senate, while Arizona has introduced a new bill (SB 1786) mandating provenance data for AI-generated media. This surge in state-level legislation occurs as the Trump administration signals its intent to challenge state AI laws it deems overly burdensome.

Sources:

Komprise Launches KAPPA to Unlock Unstructured Data with Serverless AI Compute

Komprise has introduced its AI Preparation & Process Automation (KAPPA) data services, a serverless compute platform designed to process unstructured data for AI applications. This new service empowers IT and data specialists to create and run custom data functions, such as metadata enrichment, across massive datasets without managing the underlying infrastructure. KAPPA addresses the critical challenge of leveraging vast quantities of unstructured data stored across network-attached storage (NAS), cloud, and SaaS platforms for AI. By allowing users to execute custom code on a per-file basis, the service aims to make data more searchable, governable, and valuable for training AI models. The KAPPA data services are now available to customers via an early access program.

Sources:

Comcast, Classiq, and AMD Trial Quantum Algorithms to Boost Internet Resilience

A successful trial by Comcast, Classiq, and AMD has demonstrated the practical application of quantum algorithms for improving the resilience and reliability of internet networks. The collaboration focused on the complex problem of identifying independent backup paths for network traffic during a failure—a task that grows exponentially more difficult as networks expand. Using quantum computing techniques, the companies were able to identify optimal backup routes in real-time, a critical capability for ensuring uninterrupted service. The trial’s positive results signify a key step forward in applying quantum solutions to solve real-world telecommunications challenges, moving the technology from theory to practice.

Sources:

GLM-5: New Open-Source AI Model Tackles End-to-End ‘Agentic Engineering’

The release of the new open-source model, GLM-5, represents a significant shift from large language models that generate code snippets to AI systems that can execute complete engineering tasks. This “agentic engineering” model is designed for sustained reasoning and end-to-end task completion. In complex programming environments, GLM-5’s performance is reportedly approaching that of Anthropic’s Claude Opus 4.5. The model has been scaled up to 744 billion parameters, with pre-training data expanded to 28.5 trillion tokens. It utilizes a reinforcement learning framework called Slime for large-scale training and DeepSeek Sparse Attention to maintain long-context reasoning with reduced computational cost. On benchmarks like SWE-bench-Verified, GLM-5 has achieved the highest scores among all open-source models.

Sources:

A medium-severity vulnerability has been found in the bn.js npm package, a widely used JavaScript library for big number arithmetic. The vulnerability, discovered by researcher Kr0emer, can trigger an infinite loop within the library, potentially causing denial-of-service issues. Developer security firm Snyk officially disclosed the flaw on February 19, 2026, as part of its standard procedure of working with open-source maintainers to responsibly report and address security vulnerabilities in a timely manner.

Sources: