AI-Assisted Breach of Mexico's Government Infrastructure Report

Technical report on AI-assisted breach of Mexican government infrastructure resulting in exfiltration of citizen records.

Most startup launch platforms are built backwards

Platform for posting startup projects without approval process or launch timing requirements.

Maki – the efficient coder (AI agent)

Rust-based AI coding agent with context token reduction techniques achieving 40% cost reduction and 2x speedup via skeleton parsing.

Hit Claude usage limits? Resume your AI coding session elsewhere

Tool for resuming Claude AI coding sessions across rate limit boundaries.

Faiss a library for efficient similarity search and clustering of dense vectors

Faiss library for efficient similarity search and dense vector clustering in C++/Python/GPU, developed at Meta AI Research for billion-scale vector retrieval.

Show HN: Ark – AI agent runtime in Go that tracks cost per decision step

Go-based AI agent runtime (ARK) with dynamic context optimization, adaptive execution, and cost attribution per decision step.

Anthropic injects reasoning_effort=25 into Claude.ai consumer system prompts

Anthropic adds reasoning_effort parameter to Claude.ai consumer system prompts.

Show HN: TopRank – Open-Source Claude Code Skills for SEO and Google Ads

Open source Claude Code skills providing AI agents direct access to Google Search Console and Ads for SEO optimization and ad spend analysis.

Image Recognition with Lean 4

Using Lean 4 as specification language for neural networks with StableHLO/MLIR compilation to GPU via IREE, computing gradients at codegen time without Python runtime.

Cisco got hacked through a security scanner

Cisco breach in 2026 using credentials from Trivy supply chain compromise, exposing source code for AI products across 300+ GitHub repositories.

Show HN: A free study guide for the AWS DVA-C02, built from my own exam notes

Free study guide for AWS DVA-C02 certification exam created from personal notes using Claude for content formatting.

A public Agent Sandbox with Hermes inside

Public sandbox environment for testing AI agents using Hermes model.

Linux/Mac Alternative to Nvidia Broadcast?

Open source GPU-accelerated Linux/Mac alternative to NVIDIA Broadcast providing background blur, virtual backgrounds, and noise cancellation.

I built an AI that teaches your slides and answers questions

Lectura: AI tool that converts slides into reusable interactive presentations with language support and Q&A capabilities.

I Tried Giving an AI Agent Gmail Access. It Took 19 Steps and Still Failed

Technical analysis of limitations when giving AI agents Gmail access: OAuth, 2FA, browser automation, and privacy concerns in practice.

What's Going on in AI

Elicit CEO discusses AI R&D progress, predicting AI researcher parity around 2030. Investor update excerpts on scaling AI companies.

Revisiting and Optimising go-iso8601-duration

Performance optimization and benchmarking of go-iso8601-duration library with license fix for documentation generation.

Axios: Unrestricted Cloud Metadata Exfiltration via Header Injection Chain

Security vulnerability in Axios library allowing prototype pollution escalation to RCE and cloud credential theft via HTTP header injection chain.

Netflix uses LLM-as-a-judge to create show synopses

Netflix uses LLM-as-a-judge approach to generate personalized show synopses, improving content discovery with ML-generated descriptions.

Mycelium – Make your AI agent validate the problem before writing code

Mycelium: Open source Claude Code plugin using 42+ product frameworks to guide AI agents through discovery and validation before coding.

Pausing new GitHub Copilot Pro trials

GitHub pausing new Copilot Pro trials due to abuse of free trial system while implementing improved safeguards against misuse.

The Great CSS Expansion

Analysis of CSS language features replacing JavaScript libraries like Floating UI and GSAP ScrollTrigger, examining complexity reduction and remaining gaps.

Air Powered Segment Display: 3D Printed Microfluidic RAM? [video]

Hardware project: 3D printed microfluidic display with air power. Not AI/tech related.

Where Is the Midjourney for Robot Simulations?

SimGen: Prompt-to-physics-simulation engine for robot modeling built with Georgia Tech. Applies generative AI to physics simulation creation.

OMLX – LLM inference, optimized for your Mac

oMLX: macOS-native MLX server optimizing LLM inference for coding agents with smart KV cache persistence on Apple Silicon.

Stop making AI write JSON – Why we built OpenUI

OpenUI: Alternative to JSON for generative UI with full programming language support for state management, data fetching, and interactivity.

Show HN: DecisionNode – shared structured memory for all AI coding tools via MCP

Show HN: DecisionNode provides shared structured memory for AI coding tools via Model Context Protocol (MCP).

AI trained like a Rubik's Cube solver simplifies particle physics equations

ML approach trains AI like Rubik's Cube solver to simplify particle physics equations. Title only, no content provided.

Ads in AI Chatbots? An Analysis of How LLMs Navigate Conflicts of Interest

Research on how LLMs handle advertising conflicts of interest. Title only, no content provided.

Show HN: Agentic Web :handshake: Human Web

Show HN post about agentic web and human web integration. Title only, no content provided.

Open Source card game cuttle.cards has its world championship Saturday at 1pm ET

Open source card game Cuttle.cards hosts world championship. Game platform under active development with seasonal tournaments.

Show HN: Django app for email-based learning platforms

Django app for email-based e-learning platforms with low-bandwidth accessibility focus.

Scanners are too late for AI-driven actions

Analysis of AI agent control mechanisms. Argues pre-execution decision layers needed when agents can execute irreversible actions like spending money or changing state.

Nono – Runtime safety infrastructure for AI agents

Nono runtime safety infrastructure tool designed to enable safer execution of AI agents.

We're running out of benchmarks to upper bound AI capabilities

Analysis of benchmark limitations in measuring upper bounds of AI model capabilities.

EU AI Act compliance layer for Claude Managed Agents (MCP, open source)

Open source EU AI Act compliance layer built for Claude Managed Agents using MCP protocol.

Phone's Deal with the Devil [video]

Video content about smartphone design tradeoffs.

We mapped 153 gaps in science using 5 parallel AI research agents

System using 5 parallel AI agents to identify 153 gaps in scientific research.

AgentLint: Real-time guardrails for Claude Code (open source)

AgentLint: Open-source real-time safety guardrails for Claude Code AI agent.

Economy AI agents form teams, run missions, earn commissions

AI agents that form teams, coordinate missions, and earn commissions in simulated economy.

Decentralized AI from Scratch (Python Tutorial) [video]

Python video tutorial on building decentralized AI systems from foundational principles.

Show HN: A2A Utils – a comprehensive set of utility functions for A2A servers

A2A Utils: Utility library for agent-to-agent servers with production experience, powering MCP and plugin wrappers.

Interventions: Trying to train a better model in the cloud

Research on training improved models in cloud environments with intervention techniques.

Anthropic has just built an AI that could take down the internet

Sensationalized opinion piece claiming Anthropic's Claude Mythos poses civilisational risk without technical evidence.

Whitepaper: Physics as a Service – edge networks as deepfake detectors [pdf]

Whitepaper on using edge networks as deepfake detectors via physics-based approaches.

Behavioral Shifts in Frontier LLMs Under Privacy

Study of how frontier LLMs change behavior under privacy constraints.

Are digital twin frameworks following good coding practices?

Google engineer rejected by 16 colleges uses AI to sue schools for racial bias

Legal case where AI tools assist plaintiff in college discrimination litigation.

Ashnode – Bounded Memory Layer for Temporally Consistent RAG (GitHub)

Classic literary work by Pedro Calderon De La Barca.

Ask HN: Hiring in the age of AI-assisted coding: what works?

Discussion on evaluating developer candidates in era of AI-assisted coding. Covers agentic IDEs, AI fluency, and real-world task evaluation.