AI Briefing — May 30, 2026

🔬 Research

AgentDoG 1.5: A Lightweight and Scalable Alignment Framework for AI Agent Safety and Security

97 upvotes · Dongrui Liu, Yu Li, Zhonghao Yang, Peng Wang, Guanxu Chen · Paper

A new framework for aligning AI agents in open-world scenarios, addressing safety risks from powerful execution agents like OpenClaw. The researchers trained lightweight models (0.8B-8B parameters) using just 1k samples that achieve performance comparable to GPT-5.4, while reducing deployment overhead by two orders of magnitude. This matters because existing alignment frameworks can't handle the emerging safety threats from cross-environment AI agents.

OmniRetrieval: Unified Retrieval across Heterogeneous Knowledge Sources

57 upvotes · Jinheon Baek, Soyeong Jeong, Sangwoo Park, Woongyeong Yeo, Minki Kang · Paper

A framework that takes natural language queries and automatically routes them to appropriate knowledge sources—text, tables, knowledge graphs, or property graphs—without losing each source's structural advantages. Tested across 13 datasets and 309 knowledge bases, OmniRetrieval outperforms single-source baselines by preserving the unique strengths of each data format. This addresses a major pain point in enterprise RAG systems where information is scattered across incompatible databases.

CollectionLoRA: Collecting 50 Effects in 1 LoRA via Multi-Teacher On-Policy Distillation

49 upvotes · Fangtai Wu, Hailong Guo, Shijie Huang, Jiayi Song, Yubo Huang · Paper

A method to compress up to 50 different visual effect LoRAs plus fast generation capabilities into a single LoRA for image editing. This solves the deployment nightmare of managing dozens of separate LoRA modules that interfere with each other when combined. The technique uses probabilistic routing and asymmetric prompting to maintain concept isolation while drastically reducing storage and loading costs for production image generation services.

🛠 Tools

harry0703/MoneyPrinterTurbo — AI-powered short video generation

3567 stars/day · Repo

A Python tool that generates high-quality short videos with one click using large language models. The project focuses on automating video content creation for social media platforms, leveraging AI to handle script writing, voiceover, and visual composition. This represents the growing trend of AI-first content creation tools targeting the creator economy and viral video production.

Leonxlnx/taste-skill — Anti-slop frontend framework for AI agents

2062 stars/day · Repo

A collection of agent skills that upgrade AI-built interfaces with better layout, typography, and motion instead of generic-looking UIs. The framework provides portable skills for Codex, Cursor, and Claude Code, including image-generation skills for reference boards. With 28K+ stars, this addresses a widespread frustration with the bland, cookie-cutter designs that AI coding assistants typically produce.

affaan-m/ECC — Agent performance optimization system

1406 stars/day · Repo

A comprehensive harness system for optimizing AI agent performance across Claude Code, Codex, Cursor and other platforms, featuring skills, memory, security, and research-first development tools. With 199K+ stars and active maintenance, ECC has become a critical infrastructure layer for developers building production agent systems. The project offers both universal packages and specialized agent shields for security-conscious deployments.

📡 Analysis

datasette 1.0a31

29th May 2026 · Post · tags: projects, sql, sqlite, datasette, annotated-release-notes

Datasette now supports write queries and stored queries, marking a significant evolution from read-only data exploration to full database management. Users with proper permissions can execute insert/update/delete operations through a templated interface, while the new stored queries feature allows both private and shared query collections. This transforms Datasette from a visualization tool into a complete data workspace platform.

Anthropic's run-rate revenue hits $47 billion

29th May 2026 · Post · tags: anthropic, ai

Anthropic's run-rate revenue jumped from $30 billion in April to $47 billion this month, showing unprecedented growth velocity for an enterprise software company. The numbers, disclosed in their $65B Series H announcement, represent annualized projections based on current monthly revenue and are legally verifiable since they're shared with investors. One client reportedly spent $500 million in a single month after failing to set usage limits on Claude licenses.

Sources: HuggingFace Papers API, GitHub Trending, simonwillison.net