6/30/2026
Demystifying Security Risks of AI-Powered Applications on Pre-Trained Model Hubs
Security research examining vulnerabilities in AI applications using pre-trained models from model hubs.
6/30/2026
Show HN: Trains – leaderless total-order broadcast in Rust, formally verified
TRAINS is a formally verified leaderless total-order broadcast primitive in Rust for control-plane coordination, alternative to Paxos/Raft.
6/30/2026
Accelerating LLM Inference on AMD GPUs with Low-Latency GEMMs
Technical optimization for LLM inference on AMD GPUs using low-latency GEMM implementations.
6/30/2026
Zluda (CUDA-compatible runtime for AMD) loses funding again, gains 32bit compat
ZLUDA CUDA-compatible runtime for AMD GPUs adds 32-bit compatibility and PhysX/Blender support despite losing funding.
6/30/2026
Ask HN: Where did the terminology (AI) "System Card" come from?
HN discussion thread asking about origins of AI terminology like 'system card', 'alignment', and 'safety'.
6/30/2026
Tell HN: Installing Cursor on iOS irreversibly changes your privacy settings
User reports that installing Cursor iOS app irreversibly changes privacy settings, downgrading from legacy 'Do not store code' mode.
6/30/2026
Agentic AI and the End of Static Information
Discussion of agentic AI systems replacing static information retrieval with dynamic autonomous decision-making.
6/30/2026
Embedding Forbidden Text in Spyware to Discourage AI Analysis
Security technique embedding obfuscated text in malware to complicate AI-based analysis and detection.
6/30/2026
Show HN: Makes local LLMs faster and more reliable by optimizing for your device
Tool optimizing local LLM inference with 39% faster time-to-first-token and 46% faster agent execution through adaptive resource management.
6/30/2026
Building a Jax training loop for an LLM training run
Technical guide for implementing LLM training loops using JAX framework with code examples.
6/30/2026
Show HN: An agent harness with model autorouting and memory
Minima agent harness with dynamic model routing and memory management for generalized AI task execution.
6/30/2026
Show HN: Pacific Slate – a self-hosted, model-agnostic multi-agent AI assistant
Self-hosted, model-agnostic multi-agent AI assistant supporting multiple LLM backends for deployment flexibility.
6/30/2026
AgentWire: Orchestrating many Claude Code sessions via tmux and voice-control
AgentWire is a self-hosted orchestration tool for managing multiple Claude Code agent sessions with tmux, command palette, voice control, and no telemetry.
6/30/2026
Prompt Injection Is Not a Chatbot Problem: How the Attack Surface Changes
Analysis of prompt injection vulnerabilities beyond chatbots, examining broader attack surfaces in LLM systems.
6/30/2026
Connect any self-hosted WordPress site to your AI
Plugin connecting self-hosted WordPress sites to LLM providers for content integration and AI features.
6/30/2026
Show HN: Deconstructing Anthropic's Coding Agent Control Model
Analysis of Anthropic's coding agent control model architecture showing patterns for securing autonomous agents with delegation, per-action authorization, and attribution.
6/30/2026
Useful LLM Prompts for Editing Your Own Technical Writing
Collection of LLM prompts for editing technical documentation using Claude Code for improved writing quality.
6/30/2026
Show HN: 143.dev – we open-sourced our internal coding-agent infrastructure
Open-sourced internal infrastructure for managing coding agents with team collaboration, context sharing, and observability features.
6/30/2026
Ask HN: How are you using AI agents for testings features?
Agent-QA: open source self-improving QA agent for software testing. 500M+ tokens processed, seeking community feedback.
6/30/2026
No Memory of Its Own: Governing a Visiting Agent on Sovereign Data
Title-only stub about governing AI agents operating on sovereign data. No content provided.
6/30/2026
Show HN: Trajeckt, a Firewall for AI Agents
Trajeckt: open source Rust firewall for AI agents. Enforces action plans with <5ms latency via trajectory tracking.
6/30/2026
Claude Science, an AI workbench for scientists, is now available
Anthropic launches Claude Science, an AI workbench for scientific research with MCPs and domain-specific capabilities.
6/30/2026
The Download: AI "coworkers" and stratospheric internet
Editorial on AI coworkers and workplace integration. Opinion piece from MIT Technology Review newsletter.
6/30/2026
What kinds of bugs does AI generate?
Analysis of 1,000 bugs from AI-generated code across 99 codebases, identifying patterns in agent-generated bugs and automated code review limitations.
6/30/2026
How to reverse-engineer an endpoint security product using LLMs
Using LLMs to reverse-engineer security products and discover evasion techniques for EDR/AV systems.
6/30/2026
Show HN: fenic – LLMs as dataframe operators, query meaning and structure
fenic: dataframe API with LLMs as first-class operators for unified structured/unstructured data querying.
6/30/2026
Local LLMs on a Ryzen 8700G iGPU: 13-15 tok/s on gemma4, 9-12 on qwen3.6
Benchmark results for running local LLMs on AMD Ryzen 8700G iGPU with throughput measurements.
6/30/2026
Claude Code session URL is leaking into your Git history as of v2.1.179
Claude Code session URLs leak into Git history in v2.1.179, a security issue for users of the coding agent tool.
6/30/2026
LLMs as 5x Faster Sandboxes
Performance comparison showing LLMs achieving 5x faster execution in sandbox environments.
6/30/2026
Reducing Feedback Latency with Local CI for Developers and AI Agents
Technical approach to reducing CI feedback latency for developers and AI agents using local testing.
6/30/2026
UATC – A Closed-Loop Controller to Prevent GPU OOM During LLM Training
Closed-loop controller system preventing GPU out-of-memory errors during large language model training.
6/30/2026
ChatbotX – Open-source omnichannel AI marketing platform
Open-source omnichannel AI marketing platform alternative to ManyChat and Chatfuel, supporting Claude and other LLMs.
6/30/2026
Trying Claude for E2E Testing on an Airbnb Clone
Case study using Claude AI for end-to-end testing on Airbnb clone, covering test planning, spec generation, and CI/CD integration.
6/30/2026
Start building with Nano Banana 2 Lite and Gemini Omni Flash
Google releases Gemini Nano Banana 2 Lite and Gemini Omni Flash for faster, cost-efficient image and video generation.
6/30/2026
Show HN: Evaluation Context Protocol (ECP)
ECP: Vendor-neutral protocol for portable agent evaluations across frameworks, models, and CI systems. Standardization tool.
6/30/2026
Show HN: Capacitor – shared mem for Claude Code, Cursor and other coding agents
Capacitor: Shared memory system enabling context persistence across Claude Code, Cursor and other coding agents. Open-source developer tool.
6/30/2026
Booz Allen: What's in America's Code? Testing U.S. and Chinese LLMs for Security
Security testing comparison of U.S. and Chinese LLMs by Booz Allen analyzing code vulnerabilities.
6/30/2026
Show HN: Don't ask if devs cheat with AI, test if they're good with it
Technical assessment platform evaluating candidate skills in collaborating with AI tools—prompting, reading output, debugging. Developer tool for hiring.
6/30/2026
Show HN: Let your AI agent manage your link in bio
AI agent-managed landing page builder for creators and small businesses with automated content updates.
6/30/2026
Show HN: FastReact – FastAPI and React Starter Kit for AI SaaS
FastAPI and React starter kit for AI SaaS applications with token metering and AI copilot demo.
6/30/2026
Show HN: Busabase – an approval-first database and knowledge base for AI agents
Busabase is an approval-first database and knowledge base designed for AI agents with built-in human oversight.
6/30/2026
Don't Build a Router. Train the Small Model to Know When to Defer
ML research on training smaller models to learn when to defer to larger models instead of building explicit routers.
6/30/2026
LLM-Tools – requirements.txt for agent tools
Tool management system for agent tools providing installation, versioning, discovery and execution via single configuration file for multi-agent projects.
6/30/2026
An AI trading desk built as a team of sub-agents (Claude Code and Robinhood MCP)
Multi-agent system for stock research using Claude Code and Robinhood MCP with human-in-the-loop approval workflow.
6/30/2026
Looop – A tiny, portable, Kubernetes-shaped control loop for your LLM agent
Self-contained autonomous control loop binary for agent-driven work, integrating with GitHub/Linear/Grafana and spawning worker agents via LLM decisions.
6/30/2026
Show HN: Bash is All You Need for a language model REPL
CLI-based LLM REPL built with bash and Unix tools, demonstrating minimal-dependency approach to local language model interaction.
6/30/2026
Show HN: Scaffold a BigQuery and dbt and Cube project an AI agent can operate
CLI tool for AI agents to scaffold and manage BigQuery, dbt, and Cube projects. Developer tool for data infrastructure.
6/30/2026
Show HN: Rampart on CoreML
Core ML port of Rampart PII token-classification model with Swift package. Open source ML tool for local data privacy.
6/30/2026
Show HN: Aitori – open-source tool to inspect Claude and ChatGPT traffic
Aitori: open-source traffic inspection tool for Claude and ChatGPT API interactions. Developer utility for debugging.
6/30/2026
Show HN: Research-Git, regenerate a past idea onto today's codebase
Research-Git tool captures code ideas as semantic units and regenerates them onto current codebases, compatible with Claude Code and MCP clients.