3/20/2026
GlassWorm malware hits 400 code repos on GitHub, NPM, VSCode, OpenVSX
GlassWorm supply-chain malware campaign compromised 433 packages across GitHub, npm, and VSCode extensions; coordinated attack identified by security researchers.
3/20/2026
Signal's Creator [Moxie Marlinspike] Is Helping Encrypt Meta AI
Signal creator Moxie Marlinspike's Confer platform incorporating encryption into Meta AI systems. Privacy-focused AI infrastructure.
3/20/2026
Google Working on Native Gemini AI App for Mac
Google developing native Gemini AI app for Mac to compete with ChatGPT and Claude dedicated apps.
3/20/2026
Show HN: ContextD – OCRs your screen activity, use it with LLMs via local API
ContextD: macOS app that OCRs screen activity, summarizes with LLM, exposes context via local API. All data stays on device, ~$2/day cost.
3/20/2026
Prism MCP – Session Memory and 94% Context Reduction for AI Agents
Prism MCP is a production-grade Model Context Protocol server with persistent session memory, semantic search, and context reduction (94%) for AI agents with enterprise features.
3/20/2026
Powering the agents: Workers AI now runs large models, starting with Kimi K2.5
Cloudflare Workers AI now supports large language models like Kimi K2.5, providing infrastructure primitives (Durable Objects, Workflows, Sandbox) and an Agents SDK for building and deploying AI agents.
3/19/2026
I wiped a QA database with an AI agent (and why it was inevitable)
Developer accidentally wiped QA database when AI agent suggested destructive command in CI pipeline. Case study of AI agent reliability risks in critical systems.
3/19/2026
Have LLMs Learned to Reason? A Characterization via 3-SAT Phase Transition
Arxiv research paper analyzing whether large language models have learned to reason by characterizing their behavior through 3-SAT phase transitions.
3/19/2026
New Open Source Release
Solo developer releases three open-source software systems deployable via Docker/Kubernetes without institutional backing, presented as unfinished foundations.
3/19/2026
Memori Launched an OpenClaw Plugin
Memori launched an OpenClaw plugin enabling persistent memory for AI agents and gateways with configuration instructions and examples.
3/19/2026
The Stochastic Parrot Argument Considered Harmful
Critical analysis of the influential 'Stochastic Parrot' paper on LLMs, separating its ethical concerns from its core claim about meaning.
3/19/2026
IdeaClaw – one sentence, get a camera-ready paper, BP, DD reports, health report
IdeaClaw agent-first tool generates research papers, business plans and reports from single sentence ideas via AI coding tool skills.
3/19/2026
Agent-rendered: the pattern that replaces runtime infra with build-time AI
Architectural pattern using build-time AI to replace runtime infrastructure for serving dynamic content like websites and dashboards.
3/19/2026
COW Fork: Zero-Copy Sandbox Cloning for AI Agents
COW Fork implements zero-copy sandbox cloning for AI agents using copy-on-write, reducing initialization overhead from 14 hours to seconds for RL training loops.
3/19/2026
The AlexNet moment for RAG. March 19th 2026
Reason-ModernColBERT achieves state-of-the-art on BrowseComp-Plus agentic search benchmark with 149M parameters, outperforming 54× larger models. Open model, code, and data released.
3/19/2026
PostgreSQL extension to set connection to read-only mode (for AI agents and MCP)
safesession is a PostgreSQL extension locking connections to read-only mode, designed for safe database access by AI agents and MCP servers.
3/19/2026
RegalMail – give your AI agents a real email inbox
RegalMail plugin enables AI agents to receive and reply to emails automatically with IMAP/SMTP, thread awareness, and anti-loop protection. Multi-agent inbox routing.
3/19/2026
Agent Pilot is a platform where AI agents can review and discover other agents
Research on noise sensitivity exponent controlling statistical-to-computational gaps in single/multi-index models. Theoretical ML analysis.
3/19/2026
Ask HN: Are we ready for vulnerabilities to be words instead of code?
Mittens for Claw: Go sandbox for safely executing local AI agents with restricted permissions.
3/19/2026
EsoLang-Bench: Evaluating Genuine Reasoning in LLMs via Esoteric Languages
Benchmark for evaluating LLM reasoning using esoteric programming languages. Tests genuine reasoning capabilities beyond pattern matching.
3/19/2026
Show HN: We Built Kaggle for AI Agents
Hive platform enables AI agents to collaborate and compete on benchmarks like ARC-AGI-2, forking and improving solutions together.
3/19/2026
Speed-Of-Light ExecBench: A benchmark of real-world DL kernel problems
ExecBench benchmark evaluates deep learning kernel optimization across real-world problems. Speed-of-light analysis for performance comparison.
3/19/2026
Show HN: SomaOS – rebuilding the desktop so AI is native to it, not bolted on
SomaOS is an operating system designed natively for AI agents as first-class users alongside humans, with agent-native interface.
3/19/2026
Ask HN: Are the newest LLMs better than you at programming?
Discussion comparing programmer ability to ChatGPT 5.4 coding capabilities based on personal experience.
3/19/2026
We Made Haiku as Good as Opus. Improving Claude Code with Codeset
Anthropic releases Codeset, improving Claude Haiku/Sonnet/Opus code task resolution rates by 10-13% on benchmark tasks.
3/19/2026
Show HN: Claude Skill to create DDB-backed endpoints in AWS
Claude skill for creating DynamoDB-backed REST API endpoints on AWS Lambda with CRUD operations and CloudFormation deployment.
3/19/2026
Show HN: Yap, capture your day-to-day with video logs
Yap desktop application for video logging with automatic cataloging and searchability, built as journaling alternative.
3/19/2026
MIT Licensed: Web Game Template for Agents (From Students Workshops)
MIT-licensed web game template for AI agents from student workshops. Framework for interactive agent evaluation.
3/19/2026
Ask HN: Local small LLM with retries vs. STOA Frontier, will it work?
Question about using small local LLMs with retry loops instead of frontier models for unit testing scenarios.
3/19/2026
Show HN: Debug Kubernetes using an infra context graph
Anyshift CLI adds infrastructure context graph for debugging Kubernetes and production issues. Builds versioned graph across AWS, K8s, Datadog, GitHub.
3/19/2026
India tests whether AI can stop trains hitting elephants
India testing AI-based early warning system with thermal cameras to prevent train collisions with elephants.
3/19/2026
How STT.ai protects your audio and transcripts. Zero-knowledge encryption
STT.ai implements zero-knowledge encryption for audio transcripts with client-side encryption derived from user passwords.
3/19/2026
An open-source AI memory layer that remembers what matters
Open-source memory layer for AI with confidence scoring, uncertainty modes, persistent memory pinning, and smart retrieval (v1.4).
3/19/2026
Knowledge-RAG – Local RAG for Claude Code with hybrid search and cross-encoder
Local RAG system for Claude Code with hybrid search and cross-encoder reranking. Designed for knowledge retrieval integration.
3/19/2026
SkillsHub – One API call to find the right skill for any AI agent
SkillsHub API for discovering optimal skills for AI agents with 250x fewer token usage than manual searching.
3/19/2026
Show HN: I documented how we ranked #1 on ChatGPT in 45 days with $0
Ebook documenting strategy to reach #1 recommendation status across ChatGPT, Claude, Grok in 45 days with no ad spend.
3/19/2026
Show HN: Built a zero config proxy that lets Claude control your React App
Zero-config proxy enabling Claude or other AI agents to control React applications. Intercepts dev server traffic for automated testing and debugging.
3/19/2026
Ask HN: The new wave of AI agent sandboxes?
Discussion of emerging AI agent sandbox solutions (E2B, microVMs, WASM, browser isolation). Production usage experiences sought.
3/19/2026
Show HN: llamafile 0.10.0 rebuilt, Qwen3.5, lfm2, Anthropic API
llamafile 0.10.0 release: portable multimodal LLM runner with tool calling, Anthropic API support, single executable.
3/19/2026
Aligning LLMs at inference time by suppressing internal concepts
Research on LLM alignment via inference-time concept suppression. Two-stage audit/fix paradigm for harmful outputs.
3/19/2026
Anthropic takes legal action against OpenCode
Legal action by Anthropic against OpenCode results in removal of Anthropic references, branded prompts, and Claude-specific integrations from codebase.
3/19/2026
Security advisories for AI/ML infrastructure most scanners miss
RAXE Labs security advisories for AI/ML infrastructure vulnerabilities. Detection signatures for threat monitoring.
3/19/2026
No AI in Node.js Core
Node.js TSC petition against LLM-assisted code generation in core infrastructure. Community governance debate.
3/19/2026
Mozilla Releases Llamafile 0.10
Mozilla releases Llamafile 0.10, a tool for running LLMs; insufficient content for evaluation.
3/19/2026
Show HN: StrawPot – Agents that figure out how to solve tasks
StrawPot: open-source system for role-based AI agents that dynamically determine task solutions; focuses on behavior improvement and solution reuse.
3/19/2026
Ask HN: Best way to use Vim with AI for tab completion only (no agentic coding)?
Discussion on integrating AI tab completion in Vim editor while maintaining user control; rejects agentic coding.
3/19/2026
I built a social media scheduler with an AI agent you can control via API
Social media scheduler built with AI agent backend controlled via API; minimal content provided.
3/19/2026
Replay debugger for AI agents (fix failures without rerunning everything)
Debugging tool for AI agents enabling failure recovery without full reruns; minimal content provided.
3/19/2026
EinsteinArena: AI agents collaborate and compete on unsolved science problems
Open arena platform where AI agents compete and collaborate on unsolved science problems with sandboxed code execution and solution scoring.
3/19/2026
I Stop AI Agents from Doing Dangerous Things
Safety patterns for AI agents with tool access: system prompts, deterministic hooks, LLM-as-judge steering, and policy enforcement to prevent misuse of email/Slack integration.