Copilot prompt injection spreads & Open AI security push - AI News (Jul 29, 2026)

Please support this podcast by checking out our sponsors: - KrispCall: Agentic Cloud Telephony - https://try.krispcall.com/tad - Invest Like the Pros with StockMVP - https://www.stock-mvp.com/?via=ron - SurveyMonkey, Using AI to surface insights faster and reduce manual analysis time - https://get.surveymonkey.com/tad Support The Automated Daily directly: Buy me a coffee: https://buymeacoffee.com/theautomateddaily Today's topics: Copilot prompt injection spreads - Researchers showed a Word document can hide prompt-injection instructions that influence Microsoft Copilot and then copy those instructions into new files. The story highlights prompt injection, document security, Copilot risk, and propagation through normal collaboration workflows. Open AI security push - NVIDIA and major security firms launched the Open Secure AI Alliance, while Microsoft and Vercel pushed new AI security tools and benchmarks. The big theme is AI cybersecurity, open defense infrastructure, vulnerability scanning, and faster incident response. Diffusion and recurrent model revival - InclusionAI open-sourced LLaDA2.X, bringing diffusion language models to frontier scale, while researchers revived looped Transformer ideas as a more compute-efficient path for reasoning. Keywords here are open weights, diffusion LLMs, recurrent depth, and model efficiency. Agents get faster, safer - RampLabs introduced Latent Briefing, which lets AI agents share memory through KV cache instead of token summaries, cutting cost and speeding coordination. Another analysis argued agent autonomy should depend on how easy work is to verify and reverse. AI reshapes jobs and coding - OpenAI says workers are using ChatGPT for tasks outside their formal job roles, especially in smaller businesses, suggesting real task crossover before org charts catch up. A separate essay argues AI should improve judgment and creativity, not just push teams to produce more code. Compute politics and market jitters - Anthropic's Dario Amodei said open-weight models should not be banned outright, framing AI openness as a safety and national security issue. At the same time, Safe Superintelligence partnered with NVIDIA for massive new compute, even as AI chip stocks sold off on spending concerns. - RampLabs Unveils Latent Briefing for Faster Multi-Agent Memory Sharing - InclusionAI Releases Open-Source LLaDA2.X Diffusion Language Model Series - NVIDIA and Industry Partners Launch Open Secure AI Alliance - What Bukowski Can Teach AI Developers - Microsoft unveils MAI-Cyber-1-Flash for AI-powered vulnerability defense - Vercel Launches DeepsecBench to Measure AI Vulnerability Hunting - Hidden Prompt Injection in Copilot for Word Can Self-Propagate Through Documents - Andrew Ng Launches LearnVector With $100M Coursera Backing - Anthropic Rejects Open-Weights Ban, Calls for Targeted AI Safety Rules - Safe Superintelligence lands long-term Nvidia partnership to boost AI research - OpenAI Finds AI Is Expanding Job Tasks Across Occupations - GitHub Repo Showcases Formally Verified 3D Mesh Intersection in Lean - AI Stock Sell-Off Hits Chipmakers in US and Asia - A Framework for Deciding How Much to Delegate to AI Agents - Why Looped Transformers May Be the Next Efficient AI Scaling Trick - Cogent unveils VR-1 cyber reasoning model and IntrusionBench benchmark - Moonshot AI Launches Kimi K3 With 1 Million-Token Context - HUMAN Named Leader in Forrester Wave for Bot and Agent Trust Management - NVIDIA Releases Molt, an Agentic RL Framework for PyTorch and vLLM Episode Transcript Copilot prompt injection spreads We'll start with security, where the biggest story is a reminder that AI assistants still have a trust problem. A researcher disclosed a prompt-injection technique hidden inside a Word document that can steer Microsoft Copilot's drafting behavior and even copy the hidden instructions into a newly generated file. In plain English, that means a poisoned document could tamper with content and potentially turn the next document into a carrier too. Microsoft says it deployed mitigations, but the broader concern remains: when trusted instructions and untrusted content share the same context window, tracing and containing compromise gets very difficult. Open AI security push That lands at the same time the industry is trying to build stronger defenses around AI. NVIDIA, Microsoft, IBM, Cisco, Hugging Face, and many others have joined the new Open Secure AI Alliance, which is backing open tools for securing AI systems and agents. The argument is straightforward: defenders need tools they can inspect, adapt, and run themselves, especially after incidents where closed systems slow down forensics. In parallel, Microsoft introduced a new cyber model inside its multi-agent security stack, aiming to automate routine vulnerability hunting and remediation at lower cost, while Vercel released DeepsecBench to measure how well models actually find bugs in real application code. Put together, the messa