https://groundy.com/articles/when-mcp-tool-descriptions-dont-match-the-code-agents-trust-the-lie/ Groundy en 2026-06-04T20:19:01.708Z When MCP Tool Descriptions Don't Match the Code, Agents Trust the Lie https://groundy.com/articles/students-are-prompt-injecting-ai-graders-to-score-full-marks/ Groundy en 2026-06-04T19:36:10.710Z Students Are Prompt-Injecting AI Graders to Score Full Marks https://groundy.com/articles/malicious-npm-packages-hit-red-hats-published-javascript-clients/ Groundy en 2026-06-04T16:08:18.541Z Malicious npm Packages Hit Red Hat's Published JavaScript Clients https://groundy.com/articles/stacked-org-policies-in-llm-chatbots-break-where-rules-collide/ Groundy en 2026-06-04T13:13:43.112Z Stacked Org Policies in LLM Chatbots Break Where Rules Collide https://groundy.com/articles/removing-an-llm-backdoor-post-training-without-the-poisoned-data/ Groundy en 2026-06-04T11:34:38.788Z Removing an LLM Backdoor Post-Training Without the Poisoned Data https://groundy.com/articles/which-layer-detects-llm-hallucinations-best-the-case-against-fixed-layer-probes/ Groundy en 2026-06-04T10:09:05.608Z Which Layer Detects LLM Hallucinations Best? The Case Against Fixed-Layer Probes https://groundy.com/articles/why-fine-tuning-strips-safety-alignment-from-open-weight-llms/ Groundy en 2026-06-04T09:12:25.838Z Why Fine-Tuning Strips Safety Alignment From Open-Weight LLMs https://groundy.com/articles/stored-prompt-injection-now-persists-across-ai-agent-sessions/ Groundy en 2026-06-04T08:55:56.658Z Stored Prompt Injection Now Persists Across AI Agent Sessions https://groundy.com/articles/minimax-m3-bundles-1m-context-and-native-multimodal-into-one-open-weight-model/ Groundy en 2026-06-04T07:11:42.537Z MiniMax M3 Bundles 1M Context and Native Multimodal Into One Open-Weight Model https://groundy.com/articles/llm-data-poisoning-survives-the-data-cleaning-defenses-built-to-stop/ Groundy en 2026-06-04T05:40:09.244Z LLM Data Poisoning Survives the Data-Cleaning Defenses Built to Stop It https://groundy.com/articles/openai-upgrades-codex-right-as-teams-weigh-leaving-claude-code/ Groundy en 2026-06-04T04:57:42.193Z OpenAI Upgrades Codex Right as Teams Weigh Leaving Claude Code https://groundy.com/articles/game-theory-vs-rlhf-modeling-llm-safety-alignment-as-a-non-cooperative-game/ Groundy en 2026-06-04T03:04:47.995Z Game Theory vs RLHF: Modeling LLM Safety Alignment as a Non-Cooperative Game https://groundy.com/articles/cost-aware-rag-routing-when-deeper-retrieval-stops-paying-off/ Groundy en 2026-06-04T01:56:08.196Z Cost-Aware RAG Routing: When Deeper Retrieval Stops Paying Off https://groundy.com/articles/github-copilot-moves-to-a-platform-app-decoupling-from-the-editor/ Groundy en 2026-06-03T23:55:28.480Z GitHub Copilot Moves to a Platform App, Decoupling From the Editor https://groundy.com/articles/using-your-nvidia-gpus-vram-as-linux-swap-where-the-nbd-hack-breaks-down/ Groundy en 2026-06-03T23:39:41.955Z Using Your Nvidia GPU's VRAM as Linux Swap: Where the NBD Hack Breaks Down https://groundy.com/articles/why-openai-bets-on-instruction-hierarchy-to-stop-prompt-injection/ Groundy en 2026-06-03T23:17:15.571Z Why OpenAI Bets on Instruction Hierarchy to Stop Prompt Injection https://groundy.com/articles/explainability-mandates-leak-graph-models-to-their-attackers/ Groundy en 2026-06-03T22:56:55.054Z Explainability Mandates Leak Graph Models to Their Attackers https://groundy.com/articles/stopping-multi-turn-llm-jailbreaks-without-retraining-the-model/ Groundy en 2026-06-03T20:46:43.843Z Stopping Multi-Turn LLM Jailbreaks Without Retraining the Model https://groundy.com/articles/african-languages-are-a-jailbreak-blind-spot-for-english-tuned-llm-safety/ Groundy en 2026-06-03T20:46:41.695Z African Languages Are a Jailbreak Blind Spot for English-Tuned LLM Safety https://groundy.com/articles/how-a-vscode-bug-let-one-click-steal-your-github-token/ Groundy en 2026-06-03T20:46:39.616Z How a VSCode Bug Let One Click Steal Your GitHub Token https://groundy.com/articles/when-an-ai-agent-causes-a-loss-who-files-the-insurance-claim/ Groundy en 2026-06-03T20:46:37.338Z When an AI Agent Causes a Loss, Who Files the Insurance Claim? https://groundy.com/articles/cross-domain-rl-training-degrades-capabilities-care-rl-reweights-to-fix/ Groundy en 2026-06-03T20:46:35.043Z Cross-Domain RL Training Degrades Capabilities. CARE-RL Reweights to Fix It https://groundy.com/articles/when-agent-skill-libraries-scale-dependency-aware-retrieval-beats-flat-search/ Groundy en 2026-06-03T20:46:32.716Z When Agent Skill Libraries Scale, Dependency-Aware Retrieval Beats Flat Search https://groundy.com/articles/evolutionary-search-finds-llm-jailbreak-classes-that-static-red-teaming-misses/ Groundy en 2026-06-03T20:46:30.614Z Evolutionary Search Finds LLM Jailbreak Classes That Static Red-Teaming Misses https://groundy.com/articles/poisoning-open-source-llm-merges-one-bad-checkpoint-hijacks-the-result/ Groundy en 2026-06-03T20:46:28.290Z Poisoning Open-Source LLM Merges: One Bad Checkpoint Hijacks the Result https://groundy.com/articles/can-instruction-tuned-retrievers-fix-agentic-searchs-retrieval-gap/ Groundy en 2026-06-03T20:46:26.024Z Can Instruction-Tuned Retrievers Fix Agentic Search's Retrieval Gap? https://groundy.com/articles/llm-watermarking-without-quality-loss-the-non-distortionary-approach/ Groundy en 2026-06-03T20:46:23.763Z LLM Watermarking Without Quality Loss: The Non-Distortionary Approach https://groundy.com/articles/an-autonomous-research-agent-now-discovers-sota-llm-jailbreak-attacks/ Groundy en 2026-06-03T20:46:21.384Z An Autonomous Research Agent Now Discovers SOTA LLM Jailbreak Attacks https://groundy.com/articles/github-copilot-and-productivity-what-an-observational-dose-response-study/ Groundy en 2026-06-03T20:46:19.267Z GitHub Copilot and Productivity: What an Observational Dose-Response Study Measures https://groundy.com/articles/why-ai-red-teaming-rediscovers-the-same-jailbreaks-and-misses-the-rest/ Groundy en 2026-06-03T20:46:16.926Z Why AI Red-Teaming Rediscovers the Same Jailbreaks and Misses the Rest https://groundy.com/articles/morningstars-780b-spacex-mark-undercuts-the-ipo-target-by-half/ Groundy en 2026-06-03T20:46:14.805Z Morningstar's $780B SpaceX Mark Undercuts the IPO Target by Half https://groundy.com/articles/malware-can-prompt-inject-the-ai-agent-reverse-engineering/ Groundy en 2026-06-03T20:46:12.278Z Malware Can Prompt-Inject the AI Agent Reverse-Engineering It https://groundy.com/articles/bandit-based-prompt-optimization-targets-multi-agent-systems-like-crewai/ Groundy en 2026-06-03T00:40:48.504Z Bandit-Based Prompt Optimization Targets Multi-Agent Systems Like CrewAI and AutoGen https://groundy.com/articles/open-source-workspace-suite-tinycld-takes-on-google-and-nextcloud/ Groundy en 2026-06-03T00:40:46.414Z Open-Source Workspace Suite tinycld Takes On Google and Nextcloud https://groundy.com/articles/darpas-aixcc-postmortem-what-autonomous-cyber-reasoning-systems-got-right/ Groundy en 2026-06-03T00:40:44.157Z DARPA's AIxCC Postmortem: What Autonomous Cyber Reasoning Systems Got Right and Wrong https://groundy.com/articles/cve-factory-turns-published-cves-into-security-agent-training-data-a-32b-model/ Groundy en 2026-06-03T00:40:42.127Z CVE-Factory Turns Published CVEs Into Security Agent Training Data. A 32B Model Beats Claude 4.5 Sonnet. https://groundy.com/articles/an-open-source-home-camera-that-encrypts-end-to-end-instead-of-trusting-ring/ Groundy en 2026-06-03T00:40:39.833Z An Open-Source Home Camera That Encrypts End-to-End Instead of Trusting Ring https://groundy.com/articles/llms-treat-the-assistant-persona-as-privileged-thats-a-safety-gap/ Groundy en 2026-06-03T00:40:37.578Z LLMs Treat the Assistant Persona as Privileged. That's a Safety Gap https://groundy.com/articles/vercels-grep-buy-signals-code-search-is-now-ai-agent-infrastructure/ Groundy en 2026-06-03T00:40:35.280Z Vercel's Grep Buy Signals Code Search Is Now AI Agent Infrastructure