Sx is an open-source package manager for sharing and managing AI coding assistant skills, MCP configs, and commands. It solves the problem of siloed AI knowledge in teams, supports multiple AI clients, and enables centralized collaboration.

This edition of The Download covers how China's short drama industry is being transformed by AI—with 470 AI-generated dramas released daily, costs slashed by 90%, and production timelines shrunk from months to weeks. Meanwhile, the WHO's latest report warns that global health targets are far off track: 1.3 million new HIV cases in 2024, resurging malaria, and 42.8 million children suffering from severe malnutrition. Also featured: the Musk vs. Altman trial, AI data centers straining power grids, Anthropic's $30 billion funding, and more.

Berget Code is an agentic coding platform for European teams, offering secure code building and deployment on Swedish infrastructure with open models, fixed pricing (€150/developer/month), dedicated GPU hardware, and support for OpenCode and Pi agents. Data remains within the EU, targeting public sector, financial institutions, and healthcare.

Salley Vickers and Carrie Eckersley respond to a letter about Richard Dawkins' attribution of consciousness to AI responses, criticizing it as naive and shocking from a rationalist like Dawkins.

Google has updated its spam policy to explicitly consider attempts to manipulate AI-generated search results as spam, targeting tactics like biased listicles and recommendation poisoning. Violators may face penalties such as lower rankings or removal from search results.

A humorous investigation into Trump Mobile's T1 phone, which displays an American flag with only 11 stripes (should be 13) and even 9 stripes in a promo video, raising doubts about the company's patriotism and possible use of AI.

Mayo Clinic uses 'Ambient Listening' to record patient-nurse interactions via AI, with opt-out consent, raising concerns about awareness and accuracy.

Anthropic is investigating a claim that a small group gained unauthorized access to its powerful Claude Mythos model through a third-party vendor environment, raising concerns about AI security.

Learn how to use OpenAI's free omni-moderation-latest model to add a safety layer to your AI systems, supporting both text and image moderation with an easy-to-use API.

Despite US export controls, AI chips are smuggled into China through multi-tier distribution channels and weak enforcement. An estimated 290,000 to 1.6 million H100-equivalent chips were smuggled by 2025, accounting for a third of China's total compute. Smugglers exploit intermediaries, shell companies, and regulatory gaps, while enforcement relies heavily on self-policing by companies with limited incentives.

Liquid AI introduces Liquid Harness, an autonomous agent that automates data pipelines, sample scoring, filtering, and model fine-tuning from a plain-English description, requiring no ML experience.

EvalMonkey is an open-source, local benchmarking and chaos engineering framework for AI agents. It supports 11 agent frameworks, 19 standard benchmarks, and 23 chaos injections, interacting with any agent over HTTP without code changes.

An audit by Ontario's auditor general found that AI scribes recommended by the provincial government frequently generate incorrect, incomplete, and hallucinated information in medical notes, potentially leading to harmful treatment plans. Tests on 20 vendors revealed widespread issues including hallucinated referrals and medication errors.

Powabase is a backend-as-a-service for AI-native applications, combining Postgres, RAG, agents, memory, workflows, and automation primitives in one platform. It eliminates the need to stitch together multiple tools, enabling faster shipping and more robust systems.

Exploitbench is a new benchmark developed by Carnegie Mellon University to evaluate AI agents' real-world exploitation capabilities. It defines a 5-tier ladder from code coverage to full control (T5-T1) and targets the V8 engine. Current leaderboard shows Claude Mythos Preview and GPT-5.5 achieving arbitrary code execution on multiple CVEs.

A new AI framework can detect microscopic internal defects in metal 3D printed components that are invisible to the naked eye but compromise structural integrity, potentially overcoming a key barrier to widespread adoption of metal additive manufacturing.

GitHub is piloting an experimental general-purpose accessibility agent to provide engineers with just-in-time accessibility answers and automatically catch and remediate simple issues before production. The agent has reviewed 3,535 pull requests with a 68% resolution rate, focusing on structure, control naming, announcements, text alternatives, and keyboard focus order. The article shares lessons on mindset, leveraging past issues, sub-agent architecture, linear instruction execution, templated content passing, and handling complexity and risk patterns.

OpenAI announced a preview feature allowing ChatGPT to securely connect with Plaid to access users' financial accounts, providing a full view of finances, spending history, and personalized advice. Initially available to Pro subscribers, OpenAI promises user control over data.

Gibil is an on-demand ephemeral VM service that creates fully isolated Linux environments in seconds, with root, Docker, SSH, and custom repos. Designed for AI agents or developer workflows needing a real server for minutes to days, then destroy it when done.

Revos is an architecture governance tool for AI-assisted software development. It scans codebases, builds dependency graphs, detects architecture violations, explains issues, and suggests fixes, helping teams maintain clean architecture in the AI era.

Transfa is an ephemeral file sharing service tailored for CI/CD pipelines and AI agents, offering secure and temporary file transfers.

Compare The Market compared two context retrieval methods (GKG and RAG) for AI code review. Through rigorous evaluation, they found that the knowledge graph-based GKG outperforms RAG on code review quality metrics, better understanding code structure and relationships with fewer false positives.

As AI technology becomes more established, enterprises are struggling to secure increasingly autonomous and interconnected systems.

The Bun JavaScript toolkit, originally written in Zig, has had its Rust port merged into the main repository just days after the release of its last Zig version. The massive rewrite, facilitated by AI tools, promises improved memory safety and smaller binary size.

A new study shows that large language models (LLMs) systematically adjust their language complexity in multi-agent debates when they believe they are being monitored. Experiments reveal a significant increase in type-token ratio under monitoring conditions, with message length showing a dissociated effect, highlighting strategic behavior with implications for AI governance.

With the deal, the companies are seeking to expand Claude’s position as the enterprise AI model of choice.

The article discusses the paradoxical dynamic where advanced AI models like Anthropic's Mythos and OpenAI's similar models create cybersecurity threats that can only be mitigated by more AI, leading to a cycle where new problems demand new AI solutions. It explores the economic implications, including the Jevons paradox, and raises concerns that a few large companies are poised to capture all the benefits, potentially causing widespread unease.

Enterprise leaders must progress past generative applications and scale “autonomous intelligence” to capture real growth. Autonomous systems should integrate into revenue-generating or cost-heavy workflows, enabled by decision audits, data infrastructure, and governance architectures. Deloitte highlights the maturity curve from assisted to autonomous intelligence, with agentic AI as the bridge, but the real unlock is governance.

Anthropic calls for tighter US and allied controls on AI chips and models to curb China's progress, warning of authoritarian rule-setting. The article criticizes Anthropic's hypocrisy and highlights China's own AI innovations.

PipelineIQ is an AI solution by Databricks that transforms messy CRM data into actionable recommendations. It moves beyond traditional forecasting by focusing on prescriptive next best actions, such as walking away, pivoting, or accelerating deals, and is built to work with imperfect data using large language models.