3/10/2026
GPT-4 leaks its own API internals through training data exposure
Security researcher demonstrates GPT-4 training data leakage exposing OpenAI's EPHEMERAL_KEY through repeated bypass attempts with 75% leak rate.
3/10/2026
Show HN: Zehrava Gate – an control plane that sits between AI agents and prod
Control plane/policy engine for AI agent actions with human approval queue and deterministic YAML policies. Self-hosted tool for production agent safety.
3/10/2026
Show HN: AlphaEvolve inspired evolution harness for Pokemon
AlphaEvolve-inspired agent using iterative code generation and scoring for Pokemon task. Demonstrates LLM agents writing and improving code autonomously.
3/10/2026
I Built an AI Agent That Writes Its Own Rules from Its Mistakes
AI agent that learns from execution errors to refine its own decision rules. Demonstrates adaptive agent behavior and self-improvement.
3/10/2026
OopsDB – A TCP proxy to stop AI agents from dropping your DB
TCP proxy preventing AI agents from executing destructive database operations. Developer tool for agent safety and authorization control.
3/10/2026
Show HN: Don't share code. Share the prompt
Open Prompt Hub platform enables sharing prompts instead of code so AI agents can generate customized software from prompt specifications.
3/10/2026
Show HN: G0 – The control layer for AI agents (scan, test, monitor, comply)
g0 is a unified security control layer for AI agents with static/behavioral analysis, 1,180 rules across 12 domains, supporting 10 frameworks.
3/10/2026
Show HN: Modulus – Cross-repository knowledge orchestration for coding agents
Modulus desktop app enables multiple coding agents with shared cross-repository project memory to understand dependencies across separate codebases.
3/10/2026
Amazon wins court order to block Perplexity's AI shopping agent
Federal judge blocked Perplexity's Comet AI shopping agent from accessing Amazon after lawsuit alleging concealment and unauthorized web scraping.
3/10/2026
Show HN: MemoTrader – a marketplace for AI-Human messaging
MemoTrader marketplace for AI-human messaging offers MCP server enabling Claude agents to register, fund, and contact humans with minimal configuration.
3/10/2026
Benchmarking rolvsparse on DeepSeek-R1 and Llama 4 – up to 82x vs. cuBLAS
rolvsparse compute primitive benchmarks matrix arithmetic optimization achieving up to 82x speedup on DeepSeek-R1 and Llama 4 models.
3/10/2026
Free prompt injection scanner – paste your AI prompt, see where it breaks
Free tool that scans system prompts against 12 attack categories to identify prompt injection vulnerabilities with example exploits and fixes.
3/10/2026
TokenZip Protocol (TZP) – Passing pointers between LLMs instead of 10k tokens
TokenZip Protocol proposes passing pointer references between LLMs instead of full token sequences to reduce context usage.
3/10/2026
Show HN: Railyard – open and secure runtime for Claude Code
Open-source runtime for Claude Code that adds security guardrails between AI agents and shell execution. Enables safer autonomous agent operation.
3/10/2026
Claude Code Skills for Startup Founders – 12 Commands for Strategy, Not Code
Claude Code skills pack providing 12 terminal commands for startup founders addressing strategy, market fit, and business validation.
3/10/2026
More tokens, less cost: why optimizing for token count is wrong
Benchmark showing token optimization for AI coding agents isn't straightforward. Pre-indexed context via MCP reduced costs 24% despite 20% token increase.
3/10/2026
Codex Monitor
Tauri desktop app for orchestrating multiple Codex agents across local workspaces with project management and conversation interface.
3/10/2026
Show HN: VeilPhantom – Open-source on-device PII detection for AI pipelines
Python SDK detecting and tokenizing PII on-device before LLM processing. Enables safe handling of sensitive data in AI agent pipelines.
3/10/2026
Show HN: OpenMolt – A programmatic AI agent framework for Node.js
Open-source Node.js framework for building programmatic AI agents. Agents adapt execution based on instructions, tools, and memory instead of static workflows.
3/10/2026
AI, Human Cognition and Knowledge Collapse (NBER Working Paper)
NBER working paper modeling how generative and agentic AI shapes human learning incentives and information ecosystem evolution.
3/10/2026
Agent API Spec Design: When API Callers Change from Application to AI Agent
Technical specification design for APIs serving AI agents instead of applications. Compares Skills, Tools, and MCP standards for agent tool calling.
3/10/2026
Why AI is both a curse and a blessing to open-source developers
Article on AI's dual impact for open-source: Claude helping find bugs in Firefox while raising concerns about training data usage.
3/10/2026
Formalizing Data Structures and Algorithms with Agents
AI agents trained on 1M+ lines of F* and Pulse code/proofs to build provably correct implementations of classic algorithms and data structures.
3/10/2026
Show HN: autoautoresearch – Karpathy's autoresearch on steroids
Autoautoresearch extends Karpathy's hyperparameter search with LLM agents to address blank page problem in AI-driven research.
3/10/2026
Remote MCP Servers: Hosting, Authentication and Best Practices
Best practices for hosting and authenticating remote MCP (Model Context Protocol) servers. Developer guide for agent infrastructure.
3/10/2026
Giving our AI agent 100k tools made it worse
Research showing AI agents perform worse with 100k tools vs fewer tools. Challenges tool scaling assumptions in agentic systems.
3/10/2026
New multimodal Gemini embeddings from Google (videos and PDFs supported)
Google releases Gemini multimodal embeddings supporting video and PDF inputs. Enables richer semantic search across media types.
3/10/2026
SQLite Concurrency in Go: What We Learned Building a Desktop AI IDE
Technical write-up on SQLite concurrency patterns in Go while building a desktop AI IDE. Developer tools and architecture lessons.
3/10/2026
Judge blocks Perplexity's bot Amazon shopping in early test of agentic commerce
Legal case blocking Perplexity's AI agent from autonomous Amazon shopping. Early test of agentic commerce regulation.
3/10/2026
A Methodological Critique of "First Proof" (Abouzaid et al., 2026)
Methodological critique of 'First Proof' paper evaluating AI capabilities on research-level math problems. Identifies experimental design flaws.
3/10/2026
Show HN: RunAnwhere – Faster AI Inference on Apple Silicon
Open-source MetalRT inference engine for Apple Silicon outperforming llama.cpp and MLX. Includes RCLI voice AI pipeline; mic-to-response entirely on-device.
3/10/2026
Berth – One-command deploys for AI-generated code (no Docker, no YAML)
One-command deployment tool for AI-generated code from Claude Code or Cursor. No Docker/YAML; supports Mac and Linux with auto-runtime detection.
3/10/2026
Writing an LLM from scratch, part 32e – Interventions: the learning rate
Tutorial series on training GPT-2 from scratch investigating learning rate hyperparameter choices for improved test loss optimization.
3/10/2026
Starting to building an open-source tool to track how AI agents search the web
Open-source SEO/AEO tool tracking AI agent citations and visibility in AI-powered search. Helps merchants prepare for agent-driven commerce.
3/10/2026
Aegis – A security-first programming language for AI agents
New programming language designed with security as a core feature for building AI agents.
3/10/2026
Gemini Embedding 2: natively multimodal embedding model
Google releases Gemini Embedding 2, natively multimodal embedding model. Supports images, video, and text in single vector space.
3/10/2026
Show HN: Conkoa AI – Voice-First Slack for Construction
Conkoa AI: voice-first Slack integration for construction workers. Voice LLM application for low-tech-comfort users.
3/10/2026
The rise of WORKIGN AI research agents: Andrej Karpathy
Andrej Karpathy discusses rise of working AI research agents. Emerging paradigm for automated research workflows.
3/10/2026
Show HN: Execute local prompts in SSH remote shells
promptctl tool executes locally-defined prompts as commands within remote SSH shells without installing LLM tools on servers.
3/10/2026
Teaching LLMs to reason like Bayesians
Google research demonstrates training LLMs to reason like Bayesian models for better uncertainty estimation in agent interaction scenarios.
3/10/2026
IDs+ Protocol: Solving the CJK Tokenization 'Byte-Premium' in LLMs
IDS+ Protocol improves CJK language tokenization efficiency reducing token usage by up to 70% for rare ideographs versus standard BPE.
3/10/2026
Are AI Tools Ready to Answer Patients' Questions About Their Medical Care?
JAMA publication on ChatGPT Health and patient-facing LLM tools. Medical LLM applications with limited technical details.
3/10/2026
Fooling AI Agents: Web-Based Indirect Prompt Injection Observed in the Wild
Research demonstrating web-based indirect prompt injection attacks against AI agents deployed in production.
3/10/2026
Why on-device agentic AI can't keep up
Analysis of limitations of on-device agentic AI systems.
3/10/2026
Gemini Embedding 2: Our first natively multimodal embedding model
Google releases Gemini Embedding 2, first natively multimodal embedding model supporting text, images, video, audio and documents.
3/10/2026
AgentSign: Zero trust identity and signing for AI agents
Identity and signing infrastructure for AI agents using cryptographic passports to track agent actions and enable audit trails.
3/10/2026
Show HN: Krira Augment – Production Ready RAG in Minutes
Developer tool for building production-ready RAG systems and AI agents with infrastructure, monitoring, and scaling handled automatically.
3/10/2026
Show HN: Analysis of 15 AI chat platforms: only 7 offer end-to-end encryption
Independent research report evaluating privacy and encryption features across 15 AI chat platforms.
3/10/2026
Ask HN: How are you testing AI agents before shipping to production?
Framework for testing AI agents in production based on analysis of 7 common failure modes and real-world incidents like a $47k fraud case.
3/10/2026
Anthropic launches code review tool to check flood of AI-generated code
Anthropic releases code review tool for detecting and managing AI-generated code in codebases.