Microsoft 365 E7 Gets Tools to Score, Restrict and Shut Down Rogue AI Agents
Microsoft 365 E7 is adding agent risk scores, approval gates and a kill switch for AI agents like Claude Code and Codex. Here is what IT teams get and when.
Tools & Applications
Reviews, explainers and practical guides. Which AI tools are worth using, what a new API release means for developers, and how to actually use the thing.
OpenCode Breach: Stolen API Keys Hit 1.4% of Console Users. Here Is What To Do Now
The OpenCode breach exposed provider API keys for 1.4% of Console users. Here is what happened and the steps developers should take today to stop surprise bills.
How Itaú Unibanco Is Building Customer-Facing Generative AI at Scale
Latin America’s largest private bank, Itaú Unibanco, is deploying a unified generative AI platform to serve its 70 million customers. By pairing AWS cloud architecture with strict internal governance, the institution is scaling digital-agent services while maintaining enterprise compliance and client trust.
Meta Releases Muse Spark Open: 816B Parameter Open-Weight Multimodal MoE Model
Meta Superintelligence Labs has launched Muse Spark Open, an 816-billion-parameter open-weight Mixture-of-Experts model featuring 42 billion active parameters and a 1-million-token context window. Optimized for agentic coding and native multimodal processing across text, audio, image, and video, the model introduces hardware optimizations for local and cloud infrastructure.
Black Forest Labs and Hugging Face Launch FLUX.3 Image Klein with Native Diffusers Support
Black Forest Labs has released FLUX.3 Image Klein, a 5.6-billion-parameter open-weight image generation and editing model. Arriving with day-one integration in Hugging Face’s Diffusers library, the architecture combines multi-reference image conditioning, coordinate-based bounding box layout controls, and Qwen3-VL text encoding.
When AI Agents Negotiate: The Structural Vulnerabilities of Autonomous Delegation
A satirical experiment by creator Graham Dot Zip exposing the breakdown between an AI job candidate and an AI recruiter reveals deep risks in delegation. When autonomous bots hallucinate qualifications, undercut reservation prices, and commit without human approval, simple agentic workflows quickly become financial liabilities.
Reflection AI Releases Beam: A 501B Open-Weight Sparse MoE Model for Agentic Workloads
Artificial intelligence startup Reflection AI has released Beam, an open-weight sparse Mixture-of-Experts model boasting 501 billion total parameters and 23 billion active parameters. Released under an Apache 2.0 license, Beam is optimized for code synthesis, complex reasoning, and long-horizon agentic workflows.
Phase-Separated Auditing Prevents Evidence Erasure in LLM-Assisted Formal Verification
A formal verification study by independent researcher Adarsh Agrawal demonstrates that providing correctness feedback during completeness audits in Verus inadvertently erases evidence of incomplete contracts. The paper introduces a phase-separated workflow that more than doubles certified witness discovery rates.
McKinsey Pilots AI Client Avatar Simulations in Entry-Level Consulting Interviews
Global management consultancy McKinsey & Company has introduced a 10-minute AI avatar exercise for North American entry-level applicants. Designed to simulate realistic client interactions, the tool challenges candidate reasoning while human evaluators retain full hiring authority.
The Silent Failures in Agentic Data Science: Why Autonomous Pipelines Break Without Alerts
The Mirage of HTTP 200 and Execution-Level Success In traditional software development, systems fail loudly: API calls throw exceptions, code exits with error codes, and monitoring alerts trigger immediate investigation. However, as data science teams deploy multi-agent systems—where foundation models act as reasoning engines to clean datasets, formulate hypotheses, execute SQL queries, and train
Google Launches Gemini 4 Argon for Deep Engineering, Cybersecurity, and Enterprise Workflows
Google has officially released Gemini 4 Argon, the flagship reasoning model of its new Gemini 4 generation. Built for complex, multi-step tasks across software engineering, legal research, and quantitative finance, Argon introduces expanded agentic execution and leads performance on the Vals Index.
Higgsfield AI Launches Multi-Scene System to Expand AI Video Beyond Short Clips
Generative AI platform Higgsfield has updated its workflow engine to extend AI video creation from single short-form clips into fully assembled videos running over 10 minutes. The platform leverages multi-scene stitching, model aggregation, and autonomous script orchestration.
Cameroonian HealthTech Waspito Launches AI Assistant AHIM on WhatsApp for 24/7 Triage
Africa-focused digital health platform Waspito has launched AHIM (African Health Intelligent Model), an AI-powered assistant accessible via WhatsApp. Designed to streamline clinical triage and facilitate consultations, laboratory orders, and prescription deliveries, the service aims to expand healthcare access across sub-Saharan Africa.
Sophos Achieves 96 Percent Investigation Time Cut via OpenAI Daybreak Integration
Cybersecurity provider Sophos has integrated OpenAI's Daybreak framework into its Managed Detection and Response (MDR) platform, slashing threat investigation times from 38 minutes down to 89 seconds. The AI-driven architecture now resolves 52 per cent of security cases end-to-end.
Asana slashes Browser Agent Costs 76-Fold via GPT-6.1 Sol and Prompt Cache Optimization
Enterprise work management platform Asana has achieved a 76-fold reduction in per-run model costs for its StackAI browser automation agents. By pairing OpenAI's newly released GPT-6.1 Sol model with append-only prompt caching strategies, Asana also accelerated agent execution speeds fivefold.