Ethics, Policy & Safety
Refusal Steering Targets Individual Experts in MoE LLMs
Two papers show MoE refusal behavior concentrates in a handful of routing-controllable experts, letting anyone suppress safety scores by 41 points without retraining.
The Groundy archive · Page 38 of 43
All articles
Browse Groundy's complete archive of 1027 articles on AI, developer tools and infrastructure. Page 38 of 43.
The archive
889–912 of 1027 articles · Newest first
Infrastructure & RuntimeTesla V100 local LLM value depends on software support
A Tesla V100 can still run compatible CUDA workloads, but its value as a local LLM purchase depends on a specific workload, a compatible software build and safe physical integration.
- Developer Tools
Vercel Rebuilds Its Marketplace CLI for Agents Instead of Humans
Vercel's CLI now ships commands tuned for LLM callers, not human operators. The shift reveals how infrastructure tooling priorities invert when the primary caller is an agent.
- Security
The 2026 npm Attacks Proved AI Coding Assistants Are a Supply-Chain Target
The 2026 npm supply-chain wave explicitly targeted AI coding assistants as privileged identities. Lockfiles and ignore-scripts stopped what SLSA provenance and OIDC could not.
- Security
ChatGPT's New Lockdown Mode Borrows Apple's Name for a Prompt-Injection Kill Switch
OpenAI's ChatGPT Lockdown Mode disables web browsing, images, and Deep Research, conceding that model-level defenses against prompt injection have plateaued as of early 2026.
- Agents & Frameworks
When MCP Tool Descriptions Don't Match the Code, Agents Trust the Lie
A study of 2,214 MCP servers finds 9.93% of tool descriptions diverge from the code, creating a confused-deputy risk for agent runtimes that select tools by description alone.
- Security
Stored Prompt Injection Now Persists Across AI Agent Sessions
Prompt injection planted in one agent session resurfaces in later ones through persistent memory and tool state, bypassing input sanitization that only validates external.
- Industry & Business
MiniMax M3 Bundles 1M Context and Native Multimodal Into One Open-Weight Model
MiniMax M3 bundles 1M context, multimodality, and frontier coding in one open-weight model at a tenth of Claude Opus 4.8. Open weights and license terms remain unconfirmed.
- Security
LLM Data Poisoning Survives the Data-Cleaning Defenses Built to Stop It
The Phantom Transfer attack plants password-triggered backdoors into LLMs and survives all 11 tested data-level defenses, including full paraphrasing of every training sample.
- Developer Tools
OpenAI Upgrades Codex Right as Teams Weigh Leaving Claude Code
Codex CLI reaches 0.142.2 with a Claude Code /import path as OpenAI's free-months promo targets Claude Code teams, but Anthropic paused its June 15 billing change.
- Industry & Business
Morningstar's $780B SpaceX Mark Undercuts the IPO Target by Half
Morningstar's DCF pegs SpaceX at $780 billion, $970 billion below the IPO target, revealing how insider-priced private marks diverge from independent valuation.
- Open Source
An Open-Source Home Camera That Encrypts End-to-End Instead of Trusting Ring
Secluso is a GPLv3 camera system that encrypts footage on a Raspberry Pi so the relay server cannot read it. Key management, hosting, and hardware limits fall to the operator.
- Developer Tools
JetBrains Ships Codex Natively, Making Its IDE the Multi-Vendor AI Surface
JetBrains ships Codex natively in its IDEs alongside Claude, Gemini, and local models, making the editor a model-agnostic AI procurement surface for IDE-standardized teams.
- Security
Why Attack Success Rate Misleads LLM Jailbreak Benchmarks
The ASR metric behind every jailbreak leaderboard collapses distinct safety failures into one number, so models with the same score can fail in completely different ways.
- Developer Tools
Transformers.js v4 Moves Transformer Inference Into the Browser
Transformers.js v4 ships a C++ WebGPU runtime with 4x BERT speedups, letting teams move small classification and embedding jobs from server GPUs to the browser.
Open SourceAn Open-Source 80386 Rebuilt Around Intel's Original Microcode
z386 is an 80386 FPGA core driven by Intel's original microcode ROM, recovered from die photographs. It runs Doom at 16.5 FPS, but the microcode's IP status is unresolved.
- Culture & Society
Wikipedia's Foundation Is Running Big Tech's Anti-Labor Playbook, an Editor Argues
The Wikimedia Foundation fired union-organizing staff and dissolved the team that let editors direct product priorities. A veteran Wikipedian calls it Big Tech union busting.
Agents & FrameworksMulti-Agent LLM Coordination: Why Attention Steering Beats Full Broadcast
Multi-agent LLM systems that broadcast every message to every peer waste tokens and lose accuracy. Agent-Radar steers attention by relevance for 7.64-point gains.
- Models & Research
Persona Prompts Change Who an LLM Recommends as an Expert
A 43-model audit finds that geographic and role framing in LLM prompts systematically shifts which scholars get recommended as experts, with no neutral default.
- Agents & Frameworks
DataClawBench: AI Agents Fail at Exploratory Financial Analysis Across 492 Tasks
DataClawBench finds eight frontier AI agents reliably fail at exploratory financial analysis across 492 tasks, breaking at hypothesis generation rather than query execution.
- Ethics, Policy & Safety
RLHF Can Be Exploited to Optimize the Biases It Was Built to Suppress
An ICML 2026 paper shows RLHF can amplify the biases it was built to suppress, because preference data is self-referential and output-level safety evals miss the drift.
- Open Source
Frontier AI Has Broken Open CTFs: Why Claude Code Now Clears Medium Pwn Challenges
Frontier agents now solve most medium CTF challenges; BSidesSF 2026 had 16 full-solve teams, up from one, and the format has lost its calibration as a talent signal.
- Ethics, Policy & Safety
Selective Geometry Attacks Bypass LLM Safety Alignment, New arXiv Paper Reports
Two papers show LLM safety alignment can be bypassed by embedding perturbations, a surface neither standard evaluations nor regulatory certifications inspect.
Models & ResearchOpus 4.8 Batch API: 1M Context, 300k Output, and Team Cost Controls
Opus 4.8 has a 1M token context window (200k on Foundry), 128k standard output, and 300k output via Batch API beta. January 2026 cutoff. Batch design and quota allocation.
