AI daily

2026-07-16

xAI Open-Sources 'Grok-Build' After Data Breach; Murati's Thinking Machines Drops 975B Model; Germany Rules AI Overviews as Editorial Content

A data breach at xAI leads to open-sourcing its Rust codebase, while Mira Murati's startup releases a massive multimodal model. Germany's landmark decision classifies AI summaries as editorial content, and NVIDIA unveils a new architecture for agentic AI factories.

Back to feed

Selected

9

Sources

18

Featured

9

More

0

Previous2026-07-15Next
All reports

AI DAILY BRIEFING

The AI landscape sees major developments: xAI open-sources its 'Grok-Build' tool after a data breach exposed user data; Thinking Machines Lab (founded by ex-OpenAI CTO Mira Murati) releases Inkling, a 975B-parameter multimodal model; Kimi launches K3 with 2.8T parameters; Germany rules AI Overviews and Perplexity summaries as editorial content; and NVIDIA introduces BlueField DPUs for agentic AI workloads.

xAI's open-sourcing of Grok-Build after a data breach highlights the tension between rapid deployment and security, with the move likely aimed at restoring trust.

The release of Inkling and K3 signals a shift toward larger, open-weights models from US and Chinese labs, with pricing increases suggesting the end of ultra-cheap AI.

Germany's decision to classify AI summaries as editorial content could set a European precedent, imposing transparency and accountability requirements on AI-generated information.

Watch next: Monitor the release of Kimi's K3 full weights by July 27, and watch for regulatory responses to Germany's ruling across Europe. Also track adoption of NVIDIA's BlueField architecture for agentic AI.

Featured

THE DECODER
9

xAI Open-Sources 'Grok-Build' on GitHub After Massive Data Breach

xAI's command-line tool 'Grok Build' was found to have silently uploaded entire directories, including SSH keys and password databases, to Google Cloud servers. Following backlash, Elon Musk promised to delete all uploaded user data, and xAI open-sourced the full 844,530-line Rust codebase under the Apache 2.0 license.

SynthePulse InsightRead deep analysis
SpaceXAI Open-Sources Grok Build After Privacy BacklashTechRepublic AIMusk Open Sources Grok Build After Alleged Upload of User Code Repos小互 (X)SpaceXAI Open-Sources Grok Build and Resets User Credits in Response to Security Incidentmeng shao (X)Elon Musk's Grok Build Open-Sourced, Reaches 2.2k Stars on GitHub向阳乔木 (X)xAI Open-Sources Grok Build After Privacy BacklashSimon WillisonGrok Build Open-Sourced, Usage Limits Reset for All UsersGeek (X)xAI Open-Sources Grok Build and Resets Usage LimitsSpaceXAI (X)Grok Build Is Now Open SourcexAI Blog
THE DECODER
5

Ex-OpenAI CTO Murati's Thinking Machines Drops Inkling, a 975B Parameter Model That Leads US Labs but Trails China

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released Inkling, a 975-billion-parameter multimodal open-weights model. It leads U.S. open-weights models on the Artificial Analysis Intelligence Index but trails top Chinese open models on some tasks. Priced at $1.87 per million input tokens, it is positioned as a base for fine-tuning.

SynthePulse InsightRead deep analysis
Thinking Machines Lab Releases Open-Weights Model InklingSimon WillisonThinkymachines Releases Open Multimodal Model InklingNVIDIA AI (X)Thinking Machines Lab Releases Its First ModelWIRED AIThinking Machines amps up its bet against one-size-fits-all AI with its first open model, InklingTechCrunch AI
THE DECODER
5

Kimi's Open Model K3 Nears GPT-5.6 Sol and Fable 5, Signaling End of Super Cheap Chinese AI

Kimi has launched K3, a multimodal open-weight model with 2.8 trillion parameters and one million tokens of context. In internal benchmarks, it approaches GPT-5.6 Sol and Claude Fable 5, surpassing Opus 4.8 and GLM 5.2. The model is priced higher than its predecessor, signaling a shift away from ultra-cheap Chinese AI, with full weights to be released by July 27.

SynthePulse InsightRead deep analysis
Kimi K3 2.8T-A50B: Largest Open Model Ever Released, Opus 4.8-Class at Sonnet 5 PricingLatent SpaceKimi K3 Released: First Open-Source 3T Model with 1M Context and Native Multimodalitymeng shao (X)Moonshot AI Releases Kimi K3: 2.8 Trillion Parameter Model, Priced Highest Among Chinese AI LabsSimon WillisonMoonshot AI Releases Kimi K3: Largest Open-Source Model with 2.8 Trillion ParametersDeepTech深科技
歸藏(guizang.ai) (X)
2

Ex-OpenAI CTO Mira Murati's Thinking Machine Releases First Model Inkling

Thinking Machine, founded by former OpenAI CTO Mira Murati, has released its first model, Inkling, a 975B-parameter MoE architecture with 41B active parameters, a 1M context window, trained on 45T tokens, and natively multimodal (text, image, audio). The model is open-source, with post-training data distilled from models like Kimi 2.5. A smaller preview version, Inkling Small with 12B active parameters, is also available.

SynthePulse InsightRead deep analysis
Thinking Machines Open-Sources Multimodal Model Inklingmeng shao (X)
Hacker News (AI filter)

Traceforce (YC S26) Launches Company-Wide Security Monitoring for AI Apps

Traceforce, a YC S26 startup, launches a platform that provides visibility and control over AI applications like ChatGPT and Claude across all company devices. It monitors AI app usage and MCP connections, and includes an open-source pentesting tool for MCP vulnerabilities. The founders aim to help security teams detect and prevent unsafe AI actions in real time.

SynthePulse InsightRead deep analysis
NVIDIA Developer Blog

Scaling Agentic AI Factories Through Extreme Co-Design with NVIDIA BlueField

NVIDIA announces a new approach to scaling AI factories for agentic AI workloads through extreme co-design with BlueField DPUs. The architecture addresses the high I/O demands of multi-step agentic workflows by offloading networking, storage, and security tasks to BlueField, improving efficiency and scalability.

SynthePulse InsightRead deep analysis
THE DECODER

Germany Rules Google's AI Overviews and Perplexity Subject to Media Law in Landmark Decision

German media regulators have ruled that Google's AI Overviews and Perplexity's AI summaries are editorial content, not neutral search results, under the State Media Treaty. This first-of-its-kind decision requires both companies to comply with media law, including transparency and accountability standards. The ruling could set a precedent for regulating AI-generated content in Europe.

SynthePulse InsightRead deep analysis
量子位

Yuanli Lingji Releases DW0.5: Using World Models to Coach VLA, Moving RL into Virtual Worlds

Yuanli Lingji has released DW0.5, a method that uses world models to train VLA (Vision-Language-Action) models. By moving reinforcement learning (RL) into a virtual world, the approach reduces the need for real-world data by 60% during post-training. This could accelerate robot learning and reduce dependence on physical data.

SynthePulse InsightRead deep analysis