Groundy — independent coverage of developer tools, infrastructure, and platforms
security
Stronger Safety Alignment Made LLMs Easier to Jailbreak, Not Harder
A single-query attack turns safety-trained LLMs' own refusal reasoning against them. Across 30 models, better safety judgment correlated with higher exploit rates, not lower.
securitySAML Signature Bypass Is Back: Inside the SAMLStorm Vulnerability Class
XML Signature Wrapping attacks on SAML keep recurring because the gap between validation and processing is structural. Edge WAF rules are a delaying tactic, not a fix.
policy
When LLM Safety Lives at Inference, Not Training: A Certification Gap
Post-training alignment can reshape LLM behavior after the checkpoint regulators audit, leaving a gap between the certified artifact and what actually runs in production.
cultureDo LLMs Understand Idioms in Low-Resource Languages?
MIDI tests idiom comprehension across 18 languages and finds LLMs rely on memorization over reasoning, with the sharpest failures falling on low-resource communities.
infraDoes CUDA Tile Match Hand-Tuned Kernels on Hopper and Blackwell?
CUDA Tile reaches 2.5x FlashAttention-2 on Blackwell B200 but drops to 53% on RTX PRO 6000, while Triton holds 62-101% of cuBLAS across both architectures without tuning.
securitySAMLStorm: The SAML Signature Bug That Forges Valid SSO Logins
SAML signature-confusion attacks exploit gaps between XML canonicalization and parsing, letting attackers mutate signed assertions to forge authenticated SSO sessions.
modelsMiniMax M3 Bets on Sparse Attention for 1M Context. Does the Math Hold?
MiniMax claims M3 handles 1M tokens via sparse attention, but published no technical report or independent benchmarks. Retrieval quality at full context is unverified.
modelsCan One Model Handle Every CAD Task? UniCAD Tests It
UniCAD introduces a unified benchmark and single multi-modal model for CAD reconstruction, generation, and question answering, challenging the field's per-task silos.
featured
- models Opus 4.8 vs Opus 4.7: What Changed and What Did Not
- agents Claude Code, Cursor, Copilot: How Agentic Coding Assistants Get Weaponized as Attacker Shells
- devtools Anthropic Buys Stainless: OpenAI and Google Now Depend on a Rival for SDK Tooling
- agents A New Trust Schema Exposes Why Agent Skill Registries Fail Enterprise Audit Requirements
- policy FTC's TAKE IT DOWN Act Lands May 19: 48-Hour Deepfake NCII Takedowns and No Safe Harbor
most popular
- devtools GitHub Copilot vs Cursor vs Claude Code: The 2026 AI Coding Showdown
- infra MLX vs llama.cpp on Apple Silicon: Which Runtime to Use for Local LLM Inference
- devtools Claude Code Plugins: Anthropic's Official Plugin Ecosystem Explained
- devtools GitHub Copilot's Opus 4.7 Multiplier: 7.5x to 15x to 27x in 60 Days
- models AI Code Generation Benchmarks 2026: Which Model Actually Writes Better Code?
- devtools Claude Code in GitHub Actions: A Complete Guide to Automated PR Fixes
- models Chinese AI Models Compared: DeepSeek, Qwen, Kimi, Doubao, and Ernie
- devtools GitHub Copilot Replaces Premium Request Units With Token-Metered AI Credits on June 1
- culture EU's 2027 Replaceable Battery Mandate: What It Means for Phone Buyers and Repairers Right Now
- industry Cursor's Meteoric Rise: From $300M to $3B ARR in a Year
top
feed
- jun 05 security Stronger Safety Alignment Made LLMs Easier to Jailbreak, Not Harder
- jun 05 security SAML Signature Bypass Is Back: Inside the SAMLStorm Vulnerability Class
- jun 05 policy When LLM Safety Lives at Inference, Not Training: A Certification Gap
- jun 05 culture Do LLMs Understand Idioms in Low-Resource Languages?
- jun 05 infra Does CUDA Tile Match Hand-Tuned Kernels on Hopper and Blackwell?
- jun 05 security SAMLStorm: The SAML Signature Bug That Forges Valid SSO Logins
- jun 05 models MiniMax M3 Bets on Sparse Attention for 1M Context. Does the Math Hold?
- jun 05 models Can One Model Handle Every CAD Task? UniCAD Tests It
- jun 05 models Do Foundation Models Actually Learn Relational Structure In-Context?
- jun 05 models Can LLMs Write Better Research Paper Titles Than Authors?
- jun 05 models Does Information-Theoretic Example Selection Beat kNN for In-Context Learning?
- jun 05 infra Pod-Level Remote Attestation in Kubernetes: Confidential Workloads on dstack
- jun 05 models Do Concept Bottleneck Model Benchmarks Measure Interpretability or Dataset Bias?
- jun 05 agents Cascading Hallucination in Agentic RAG: When One Bad Retrieval Poisons the Chain
- jun 05 security Vercel's Flags SDK Exposed Feature-Flag Definitions via CVE-2025-46332
- jun 05 models Continuous Bit-Width Quantization vs Fixed INT4: Does LiftQuant Beat Discrete?
- jun 04 models Federated Learning for Industrial IoT Anomaly Detection: The Data-Locality Tradeoff
- jun 04 infra Generating GPU Kernels for Moore Threads Silicon: Can LLMs Break CUDA Lock-In?
- jun 04 devtools Alibaba's Open Code Review Moves AI Review Into the CLI, Not the PR
- jun 04 infra Microsoft's Azure Linux Goes General-Purpose: The Container Base-Image Play
- jun 04 models Reading Failed LLM Reasoning Traces Won't Tell You Which Ones RL Can Fix
- jun 04 agents Can AI Agents Build Other Agents? The Meta-Agent Challenge Says Mostly Not Yet
- jun 04 models Can You Stitch Two Foundation Models Together Without Retraining?
- jun 04 infra Cloudflare Acquires VoidZero, the Company Behind Vite's Rust Toolchain
- jun 04 security Jailbreak Suffixes Hit Harder at Specific Token Positions, New GCG Variant Shows
- jun 04 policy When Should an LLM Forget You? A Benchmark for Deciding What Memory to Drop
- jun 04 security OpenAI Adds Lockdown Mode to ChatGPT, Shifting Prompt-Injection Risk to Users
- jun 04 policy When RL Training Rewards Capability-Seeking: A New Alignment Risk
- jun 04 models Do Reasoning LLMs Waste Tokens? OckBench Tries to Measure It
- jun 04 security Activation Steering Was Sold as LLM Control. New Work Makes It an Attack Surface
- jun 04 culture Can Teaching Logical Fallacies Inoculate People Against AI Misinformation?
- jun 04 devtools Vercel Ships Experimental Native CLI Binaries to Cut the Node Startup Tax
- jun 04 security Catching LLM Agents Leaking Credentials From Their Own Activations
- jun 04 policy Refusal Steering Targets Individual Experts in MoE LLMs
- jun 04 infra Putting a Datacenter V100 in a Gaming PC: The Local LLM Math
- jun 04 devtools Vercel Rebuilds Its Marketplace CLI for Agents Instead of Humans
- jun 04 security The 2026 npm Attacks Proved AI Coding Assistants Are a Supply-Chain Target
- jun 03 security ChatGPT's New Lockdown Mode Borrows Apple's Name for a Prompt-Injection Kill Switch
- jun 03 agents When MCP Tool Descriptions Don't Match the Code, Agents Trust the Lie
- jun 03 security Students Are Prompt-Injecting AI Graders to Score Full Marks
- jun 03 devtools Malicious npm Packages Hit Red Hat's Published JavaScript Clients
- jun 03 policy Stacked Org Policies in LLM Chatbots Break Where Rules Collide
- jun 03 security Removing an LLM Backdoor Post-Training Without the Poisoned Data
- jun 03 models Which Layer Detects LLM Hallucinations Best? The Case Against Fixed-Layer Probes
- jun 03 policy Why Fine-Tuning Strips Safety Alignment From Open-Weight LLMs
- jun 03 security Stored Prompt Injection Now Persists Across AI Agent Sessions
- jun 03 industry MiniMax M3 Bundles 1M Context and Native Multimodal Into One Open-Weight Model
- jun 03 security LLM Data Poisoning Survives the Data-Cleaning Defenses Built to Stop It
- jun 03 devtools OpenAI Upgrades Codex Right as Teams Weigh Leaving Claude Code
- jun 03 policy Game Theory vs RLHF: Modeling LLM Safety Alignment as a Non-Cooperative Game