 | Bardic Labs AI solutions and automations, built in Singapore. Bardic Labs builds practical AI systems for teams across Singapore and Southeast Asia: document pipelines, internal copilots, and customer-facing agents that run in production rather than in a demo. Start with a free automation audit and find out what is worth handing to a machine. Book a free audit → |
OpenAI hits the brakes on its own AI race just as the US-China gap narrows and rival labs ship one always-on agent after another.
OpenAI says it slowed some AI development to shore up security, even as it races to match Anthropic on enterprise data privacy ahead of a possible IPO. Meanwhile, a new report says the US lead over China in AI is shrinking fast, and a separate assessment finds that no major AI lab fully applies basic safety controls to its own internal systems. On the product side, Anthropic, xAI, DeepSeek, Meta, and Cerebras all shipped new agents, apps, or hardware, while open-source tools like Block's Berd and Unsloth's quantization format kept the community buzzing. Security agencies also warned that attackers are now using AI to build exploits against industrial control systems.
| 1 | theverge.com OpenAI Hits the Brakes on AI DevelopmentOpenAI said it slowed the pace of some AI development, including a two-week pause, to tighten security and safeguards even as it faces pressure from a looming IPO and rivals like Anthropic. |
| 2 | techcrunch.com OpenAI Ups the Ante on Enterprise Privacy to Match AnthropicOpenAI is rolling out new privacy protections for enterprise customer data as it competes with Anthropic to be seen as the safest place for businesses to store their information. |
| 3 | news.google.com US Lead in AI Race With China Is Rapidly NarrowingA new report finds the gap between the US and China in artificial intelligence is closing faster than expected, raising fresh questions about America's technological edge. |
OpenAI confirmed it will keep Zero Data Retention available for eligible API customers and previewed 'Private Safety Processing,' a way to run advanced AI safety checks without compromising data privacy.
Read →Anthropic's Claude assistant gained the ability to send emails through Gmail on a user's behalf without requiring confirmation before each send.
Read →Cerebras announced its new CS-4 system, drawing heavy attention on Hacker News as the company continues to push wafer-scale chips as an alternative to GPU-based AI computing.
Read →xAI rolled out Grok Bot, a persistent AI agent designed to run continuously rather than only responding when prompted.
Read →DeepSeek released open-source software described as filling a missing piece of infrastructure between AI models and the agents that use them.
Read →Meta's AI chatbot is now available as a standalone Mac app that supports screen sharing, so the AI can react to what's on your screen, plus system-wide dictation.
Read →
Z.ai's GLM-5.3 scored 60 points on the Artificial Analysis Intelligence Index, tying Kimi K3 for the top spot among open models and beating its predecessor GLM-5.2 by seven points, though its full release has been pushed back.
Read →The NSA, CISA, and FBI say hackers are using AI to write exploit scripts targeting Siemens S7 industrial controllers, sharply lowering the skill and time needed to attack critical infrastructure like energy and water systems.
Read →A widely discussed paper, trending on Hacker News, examines how AI systems are changing the practice and culture of mathematics.
Read →In a podcast conversation highlighted by developer Simon Willison, the case is made that measuring coding productivity in lines of code makes more sense now that AI agents can produce far more debugged code per day, shifting the real bottleneck to human cognitive capacity.
Read →Researchers re-evaluated memory-based self-improving agents, which learn from a stream of tasks by keeping a text memory bank, and found their performance is highly noisy and depends heavily on the order tasks are presented in.
Read →A new framework called TokEval evaluates properties of language model tokenizers—the systems that break text into chunks—and finds specific properties predict downstream abilities like math reasoning and code generation.
Read →
Block, the company behind Square and Cash App, has open-sourced Berd, an Apache 2.0-licensed desktop app that lets users work with AI agents across different models, tools, and local files and projects.
Read →Unsloth's new Dynamic 3.0 GGUF quantization format drew significant Hacker News attention, offering a way to shrink AI models for local use while preserving performance.
Read →The Ornith-1.5 project, trending on Hacker News, describes an approach for AI agents to progress from building their own scaffolding to genuinely improving themselves over time.
Read →A YC-backed team launched OneCLI, an open-source tool that gives teams a sandboxed environment for safely running AI coding agents.
Read →Liquid AI published quantized checkpoints of its LFM2.5 model created through quantization-aware distillation, a technique meant to shrink models while retaining more of their original accuracy.
Read →DFlash 2, trending on Hacker News, presents a new approach to keeping multiple drafts generated in parallel, aimed at speeding up AI text generation workflows.
Read →The AI industry, condensed into a five minute read. Free, and you can leave whenever.
Subscribe free