Ethics, Policy & Safety
Do LLM Personality Tests Measure Anything? A New Paper Says No
A June 2026 arXiv preprint finds 81 to 90 percent of LLM personality test variation stems from directional response bias, undermining persona and safety scores.
The Groundy archive · Page 27 of 38
All articles
Browse Groundy's complete archive of 896 articles on AI, developer tools and infrastructure. Page 27 of 38.
The archive
625–648 of 896 articles · Newest first
- Security
Reported React Server Components Leak Is Unconfirmed: Audit the Payload
A reported React Server Components source-code leak has no CVE or advisory in React or Vercel's channels. Audit what your app serializes before trusting the boundary.
- Developer Tools
Generating Vercel Firewall Rules From Natural Language: What to Audit
Vercel's natural-language WAF rules silently fill rate-limit defaults and persistent block durations. Read the generated config before publish, not the prompt.
- Developer Tools
GLM-5.2 Coding Plan vs Claude Opus 4.8: Picking a Model for Coding Agents
GLM-5.2 ships an MIT-licensed Coding Plan at $12.6 to $112 per month, forcing coding-agent teams to weigh billing model and license terms over the missing benchmark table.
- Security
Vercel's Secure AI Agent Guidance Pushes Defense Into the Sandbox
Vercel treats prompt injection and agent hallucination as unsolvable at the model layer, routing defense into per-session sandboxes and shifting security onto deployment ops.
- Security
Nx Supply-Chain Attack Used Developers' Own AI CLIs to Hunt Secrets
s1ngularity's malicious Nx packages invoked installed AI CLIs with permission bypass flags to enumerate secrets, making any local agent a scriptable recon primitive.
- Industry & Business
Vercel Folds Backends, Agent Tooling, and Operations Into Its Deploy Platform
At Ship 2026, Vercel launched an agentic infrastructure platform that folds backends, agent tooling, and operations into its deploy stack, raising switching costs.
- Infrastructure & Runtime
Cloudflare Now Routes Public Traffic to Private Apps via DNS, No VPN
Cloudflare's private-origins DNS routing lets public hostnames reach RFC 1918 apps without a VPN, but the flag only routes traffic; it does not authenticate callers.
- Open Source
OpenAI's Patch the Planet Is Security Capacity for Nine Projects, Not Sustainability Funding
OpenAI's Patch the Planet routes AI security research, ChatGPT Pro, and API credits to nine named projects, leaving the maintainer triage economics beneath them untouched.
- Open Source
MiniMax M3 Claims GPT-5.5-Beating Code With 1M Context and Open Weights
MiniMax M3 bundles open weights, a 1M-token context, and multimodal input, and claims a 0.4-point coding edge over GPT-5.5 that nobody has verified.
- Industry & Business
George Hotz Says Only AGI Doom Justifies Today's AI Valuations
Hotz argues OpenAI's $500B valuation only closes in a DCF model via an AGI assumption, fusing the safety narrative and investor pitch into a timeline no incumbent can revise.
- Infrastructure & Runtime
GitHub's AI Capacity Crunch Pushes Microsoft to Rent AWS Compute
Microsoft reportedly rents AWS compute to keep GitHub's AI inference running after Azure fell behind, signaling that owned infrastructure no longer self-supplies AI load.
- Ethics, Policy & Safety
Community LoRA Mining Raises a Consent Gap for Style Generation
FreeStyle, a June 2026 preprint, mines community LoRA adapters as training data for image generation, shifting licensing burden onto contributors and platforms like Civitai.
- Culture & Society
Why Audio Deepfake Detectors Keep Losing the Voice-Cloning Arms Race
A 34,000-parameter audio deepfake detector reaches only 75 to 80 percent cross-domain accuracy, a result that shows why post-hoc detection sits downstream of generation.
- Security
Mixed Compliance Data Makes Safety Fine-Tuning a Curation Problem
A June 2026 preprint shows benign and harmful compliance examples are not interchangeable, with DPO, not SFT, the stage that stops benign examples from amplifying harm.
- Ethics, Policy & Safety
When an LLM Narrates a Solver, the Explanation Drifts From the Math
A June 2026 arXiv paper isolates the narration gap in LLM-solver loops: prompt injection can invert a verified verdict at the prose stage, breaking reasoning-log audits.
- Infrastructure & Runtime
Cloudflare's Temporary Accounts Give AI Agents Disposable Credentials
Cloudflare's temporary accounts give agents auto-expiring 60-minute credentials, but the launch is an onboarding shortcut, not a scoped security control.
- Ethics, Policy & Safety
Grading DiffusionGemma: How an Open-Weight Diffusion Model Scores on Transparency
An arXiv paper finds DiffusionGemma's opaque serial depth collapses from 28.6x to 1.1x via a token bottleneck, though its model card leaves training data unitemized.
- Ethics, Policy & Safety
Who Owns Editorial Authority When LLMs Mediate Knowledge?
A June 2026 preprint argues no role in the LLM pipeline holds editorial sign-off for what answer engines surface as public knowledge, framing it as a governance gap.
- Open Source
Lithuania's Open-Source Drone-Detection Network Signals an Air-Defense Shift
A Lithuanian open-source drone-detection network points to cheap passive sensor meshes that could move air defense off centralized radar, though unvalidated by field tests.
- Culture & Society
Why AI Misreads Nigerian English: A Register Gap in Public Discourse
Models tuned on standard English misread Nigerian English and Pidgin register shifts, pushing intent validation onto local annotators vendors rarely fund.
- Agents & Frameworks
Deep-Research Benchmarks Hide How Agents Fail at Open-Web Source Grounding
June 2026 benchmarks like DRFLOW show agents that ace curated RAG retrieval still fail to ground primary sources in the open web, leaving citation audits to humans.
- Ethics, Policy & Safety
Vector Database Access Control Is Missing, and RAG Pipelines Pay for It
Production vector databases enforce access control at the collection boundary, not per embedding, so RAG retrieval can leak chunks a user's row-level policy blocked.
- Agents & Frameworks
DSPy Ships Autonomous Prompt Optimization, but Judge Drift Is the Failure Mode
DSPy's GEPA optimizer already tunes LLM prompts with no human in the loop, but judge drift and trajectory collapse are the failure modes when the metric is wrong.