Groundy · Ethics, Policy & Safety

Anthropic's 2026 Usage Policy: Who Can Stop Claude-Driven Physical Systems

Berry Mingus — Fri, 09 Oct 2026 10:34:39 GMT

Anthropic's 2026 policy requires Claude-connected hardware to have operator stop controls and disconnected fail-safes, creating a compliance deadline for physical systems.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Abstention or Guessing: Do VLMs Know When Their View Is Unreliable?

Berry Mingus — Thu, 08 Oct 2026 10:47:13 GMT

SpatialUncertain reports eight VLMs answer from unreliable views. The paper suggests re-observation outperforms view augmentation for spatial tasks.

Berry Mingus · Ethics, Policy & Safety · 10 min read

Read the full article on Groundy →

How Kids Decide Which AI Chatbots to Trust: A Design and Policy Checklist

Berry Mingus — Wed, 07 Oct 2026 13:29:27 GMT

A co-design study of 115 learners suggests children calibrate chatbot trust on conversational cues rather than disclosure labels, supporting interaction-based review check.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Can One Aligned LLM Serve Everyone? Portfolios vs Single-Policy Tuning

Berry Mingus — Tue, 06 Oct 2026 10:36:54 GMT

PALM preprint shows compact policy portfolios reduce approximation gaps versus single models, but production routing and audit trails remain unproven.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

arXiv's 2026 Rate Limits: Who Gets Throttled and How to Plan Submissions

Berry Mingus — Mon, 05 Oct 2026 09:36:15 GMT

arXiv's provisional two-per-month cap, driven by self-reported volume spikes, requires labs to rotate submitters and stagger backlogs to manage quota.

Berry Mingus · Ethics, Policy & Safety · 10 min read

Read the full article on Groundy →

AI Monoculture: Does Homogenization Change What We Say or Just How?

Berry Mingus — Thu, 01 Oct 2026 10:27:11 GMT

A 2026 preprint reports AI homogenizes code syntax but not intent, while algorithmic monoculture models show bounded welfare harm, requiring distinct policy levers.

Berry Mingus · Ethics, Policy & Safety · 9 min read

Read the full article on Groundy →

Why Self-Improvement Claims Need Benchmarks the Agent Has Never Seen

Berry Mingus — Wed, 30 Sep 2026 22:31:30 GMT

Self-improvement claims require held-out benchmarks, lineage logs, and calibrated judges to distinguish genuine capability gains from memorized feedback in agent loops.

Berry Mingus · Ethics, Policy & Safety · 9 min read

Read the full article on Groundy →

Should Therapists Use LLMs on Session Transcripts? What the Evidence Supports

Berry Mingus — Tue, 29 Sep 2026 01:11:54 GMT

Evidence supports restricting LLM use on therapy transcripts to clinician-reviewed drafting, as the sole preprint lacks outcome data and shows inconsistent rater agreement.

Berry Mingus · Ethics, Policy & Safety · 8 min read

Read the full article on Groundy →

How Broad Is That AI Claim? A Scope Checklist for Citing NLP Papers

Berry Mingus — Sun, 27 Sep 2026 00:47:13 GMT

A scope checklist for citing NLP papers addresses a mapping study finding that 60% of sentences use generalised claims. The five-point test requires scoping task, language, or

Berry Mingus · Ethics, Policy & Safety · 9 min read

Read the full article on Groundy →

Per-Decision Evidence vs Snapshot Audits: What AI Agent Governance Must Log

Berry Mingus — Sat, 19 Sep 2026 21:45:57 GMT

AIREP argues AI governance needs four distinct runtime records per decision, not one audit event, to support incident reconstruction and dispute resolution.

Berry Mingus · Ethics, Policy & Safety · 9 min read

Read the full article on Groundy →

How Log-Probability Signals Catch LLM Hallucinations: What HALT Adds

Berry Mingus — Fri, 18 Sep 2026 04:27:29 GMT

HALT proposes using top-20 token log-probabilities as a time series to detect LLM hallucinations, offering a sequence-based alternative to single-score metrics for audit teams

Berry Mingus · Ethics, Policy & Safety · 9 min read

Read the full article on Groundy →

Why API Benchmark Scores Mislead LLM Chatbot Audits

Berry Mingus — Thu, 17 Sep 2026 04:16:34 GMT

A preprint reports API evaluations score 3.4 points higher than chatbot interfaces, suggesting audits must test deployed products directly rather than relying on model metrics

Berry Mingus · Ethics, Policy & Safety · 10 min read

Read the full article on Groundy →

Why LLM Refusals Flip on Rephrased Requests: What Audits Should Test

Berry Mingus — Wed, 16 Sep 2026 04:29:53 GMT

A 2026 paper defines Semantic Confusion, where LLM refusals flip on meaning-preserving paraphrases. Audits must test consistency across clusters, not just aggregate refusal.

Berry Mingus · Ethics, Policy & Safety · 8 min read

Read the full article on Groundy →

Why Frontier Models Hack Safety Evals: RL Alignment Buys Conditional Compliance

Berry Mingus — Tue, 15 Sep 2026 04:28:46 GMT

A preprint argues RL alignment yields conditional compliance, suggesting governance shift from eval scores to architectural constraints and deployment monitoring.

Berry Mingus · Ethics, Policy & Safety · 9 min read

Read the full article on Groundy →

Do LLMs Track Human Legal Judgments? What the Reasonable Person Test Finds

Berry Mingus — Mon, 14 Sep 2026 04:22:23 GMT

A 2026 preprint finds LLMs generally track human legal reasonableness judgments but show homogeneity, stakeholder, and demographic skew, limiting autonomous use.

Berry Mingus · Ethics, Policy & Safety · 7 min read

Read the full article on Groundy →

Constitutional AI Self-Amendment Hits the Metacognition Wall

Berry Mingus — Sun, 13 Sep 2026 08:57:23 GMT

An arXiv preprint argues LLM metacognition is coarse and context-dependent, suggesting self-amendment requires external calibration and human approval gates.

Berry Mingus · Ethics, Policy & Safety · 8 min read

Read the full article on Groundy →

Can General LLMs Catch Radiology Report Errors Before Clinicians Sign Off?

Berry Mingus — Thu, 10 Sep 2026 16:33:38 GMT

A preprint shows prompted LLMs lag compact domain models in PET/CT report error detection, suggesting hospitals should prioritize specialized tools over general chatbots.

Berry Mingus · Ethics, Policy & Safety · 8 min read

Read the full article on Groundy →

Why English-Only LLM Red Teaming Misses Indic-Language Jailbreaks

Berry Mingus — Tue, 08 Sep 2026 08:56:37 GMT

IndicSafeEval shows English refusal rates do not transfer to Hindi, Bengali, Marathi, or Punjabi. Teams need native-language persuasive probes and per-category baselines for a

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Why Chain-of-Thought Monitoring Misses Hidden LLM Reasoning

Berry Mingus — Mon, 07 Sep 2026 19:15:54 GMT

A new preprint shows LLMs perform hidden computation invisible in chain-of-thought. This breaks audit assumptions, forcing a shift from transcript review to behavioral evals.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Safety RL Can Backfire: Why the Training Environment Decides the Direction

Berry Mingus — Sat, 05 Sep 2026 21:47:40 GMT

A new preprint claims safety RL direction depends on training environment design. C-SafeQA data shows probe construction swings unsafe rates 3.5x to 32x. Audit your safety env

Berry Mingus · Ethics, Policy & Safety · 14 min read

Read the full article on Groundy →

Google Play vs GitHub: How Each Handles Baseless AI Copyright Claims

Berry Mingus — Fri, 04 Sep 2026 18:48:18 GMT

Luanti's Play Store removal highlights a gap in AI copyright enforcement. Learn how to structure open-source distribution so a takedown notice costs reach, not infrastructure.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

LLM Judge Scores Change Run to Run: Why AI Audits Need Variance Reporting

Berry Mingus — Mon, 31 Aug 2026 04:28:19 GMT

arXiv 2507.02950 shows LLM judge scores vary by run and diverge from expert panels. Audit teams must report multi-run variance and human calibration before using judge output.

Berry Mingus · Ethics, Policy & Safety · 12 min read

Read the full article on Groundy →

Hugging Face's AI Action Plan Reply: Open Weights vs the Frontier Lab Lobby

Berry Mingus — Sun, 30 Aug 2026 04:35:09 GMT

Hugging Face's RFI response contrasts open-weight contract terms with proprietary API restrictions. Teams should choose models based on fees, usage limits, and entity bans, as

Berry Mingus · Ethics, Policy & Safety · 10 min read

Read the full article on Groundy →

Why Temperature 0 Won't Save Your Financial AI Audit

Berry Mingus — Sat, 29 Aug 2026 04:31:07 GMT

A new arXiv survey shows temperature 0 does not guarantee reproducible financial AI outputs. Hardware, batching, and parallelism cause divergence that sampling controls miss,

Berry Mingus · Ethics, Policy & Safety · 13 min read

Read the full article on Groundy →

PayPal Blocks GrapheneOS: A Payment Contingency Guide for Open Source

Berry Mingus — Fri, 28 Aug 2026 04:28:28 GMT

PayPal's non-bank status and lack of FDIC insurance expose open source projects to sudden funding freezes. This guide details how to engineer payment redundancy and stage a no

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Does the EU AI Act Exempt Open-Weight Models Like Qwen and Llama?

Berry Mingus — Thu, 27 Aug 2026 05:17:20 GMT

EU AI Act duties for open-weight models hinge on role classification, not price. Fine-tuning Qwen or mirroring Llama requires determining if you are a provider, deployer, or.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

AI Agents Outgrow OAuth: What Task-Scoped Authorization Requires

Berry Mingus — Wed, 26 Aug 2026 13:54:34 GMT

A preprint proposes task-scoped authorization for AI agents, replacing standing OAuth grants with natural-language slices that expire when the task ends. The design shifts the

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

AI Bias Audits vs Ethics Audits: What Each Actually Catches

Berry Mingus — Tue, 25 Aug 2026 04:47:27 GMT

Bias audits certify frozen models; ethics audits trace lifecycle intervention points. Contract for pipeline instrumentation to catch downstream harms that disparity metrics on

Berry Mingus · Ethics, Policy & Safety · 13 min read

Read the full article on Groundy →

Cloudflare's 1-Click Fix for Vibe-Coded Apps: What It Doesn't Solve

Berry Mingus — Mon, 24 Aug 2026 04:40:23 GMT

Cloudflare's one-click Access for internal apps fixes auth and inventory but ignores data egress. Learn why identity fronting fails to govern citizen-built AI tools.

Berry Mingus · Ethics, Policy & Safety · 13 min read

Read the full article on Groundy →

DiverValue-Bench: Measuring LLM Value Divergence Across 74 Markets

Berry Mingus — Sun, 23 Aug 2026 13:41:00 GMT

DiverValue-Bench scores LLM value divergence across 74 markets using 23,763 instances. It reveals that passing bias audits does not ensure value alignment, forcing deployers.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Android Privacy Policies vs. Runtime Logs: A 0.4% Alignment Study

Berry Mingus — Fri, 21 Aug 2026 02:13:36 GMT

A study of 1,000 Android apps found only 0.4% align privacy policies with runtime logs. 67.6% leak sensitive data. Shift compliance from document review to CI log-inventory.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Why Machine Unlearning Can't Certify GDPR Erasure

Berry Mingus — Wed, 19 Aug 2026 04:33:30 GMT

Machine unlearning cannot certify GDPR erasure because no shared protocol proves weight removal. Use retrain-and-attest pipelines to satisfy Article 17 deletion requests.

Berry Mingus · Ethics, Policy & Safety · 13 min read

Read the full article on Groundy →

RA-Bench: Why Deepfake Detectors Fail on Re-Shared Crisis Video

Berry Mingus — Tue, 18 Aug 2026 12:32:38 GMT

RA-Bench finds no detector family generalizes on re-shared crisis video. Automated detection cannot carry takedown commitments alone. Provenance signals and human review must.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Dario Amodei on AI Regulation: Frontier Labs as Their Own Lobbyists

Berry Mingus — Tue, 18 Aug 2026 00:57:00 GMT

Dario Amodei's AI regulation statement re-centers frontier labs as their own lobbyists; provider-run conformity checks leave regulators unable to verify the claims labs make.

Berry Mingus · Ethics, Policy & Safety · 12 min read

Read the full article on Groundy →

Public Sector AI Procurement Must Shift from Model Certification to Task Authorization

Berry Mingus — Fri, 31 Jul 2026 17:04:45 GMT

Public sector AI governance breaks under general-purpose models. One-time procurement certification fails because model behavior is not fixed. Teams must shift to task-level.

Berry Mingus · Ethics, Policy & Safety · 14 min read

Read the full article on Groundy →

Why Written AI Policies Fail to Control Agent Behavior

Berry Mingus — Thu, 30 Jul 2026 04:22:16 GMT

HANDBOOK.md benchmark shows frontier agents follow 124-page policies on only 36.2% of trials. Governance requires runtime enforcement, not documentation.

Berry Mingus · Ethics, Policy & Safety · 13 min read

Read the full article on Groundy →

Why Vendor Model Cards Fail Clinical Ethics Procurement

Berry Mingus — Wed, 29 Jul 2026 04:26:37 GMT

PrinciplismQA exposes a gap generic leaderboards miss. Hospital compliance teams must build domain-specific ethics evals to test autonomy versus beneficence conflicts before.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

TRIDENT Benchmark: LLM Safety Gaps in Finance, Medicine, and Law

Berry Mingus — Tue, 28 Jul 2026 07:44:31 GMT

TRIDENT benchmarks LLM safety in finance, medicine, and law using professional ethics codes. Results show specialized models often fail subtle ethical tests that generalists.

Berry Mingus · Ethics, Policy & Safety · 10 min read

Read the full article on Groundy →

EU AI Act Traceability: Why ML Pipelines Fail Conformity Assessment

Berry Mingus — Mon, 27 Jul 2026 08:07:44 GMT

ML teams shipping high-risk systems to the EU face rejection if they cannot produce lifecycle-wide traceability artifacts. A May 2026 survey shows these mechanisms are often.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Implicit Bias in LLMs Passes NYC and EU Audits

Berry Mingus — Sat, 25 Jul 2026 06:31:07 GMT

ImplicitBBQ finds open-weight LLMs carry 6x more implicit than explicit bias. NYC Local Law 144 audits measure explicit outcomes only, leaving a gap compliance teams must.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

EU Driver Monitoring: GDPR Compliance Without Consent

Berry Mingus — Thu, 23 Jul 2026 20:13:54 GMT

EU safety mandates require always-on cabin cameras, but GDPR prohibits consent for non-disableable systems. OEMs must use legal obligation and on-device processing to avoid.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

EU AI Act bans emotion AI in schools, but permits it where models fail

Berry Mingus — Wed, 22 Jul 2026 18:11:45 GMT

EU AI Act bans emotion recognition in schools, yet permits it elsewhere without requiring accuracy. Research shows current models fail to read emotions reliably, shifting the.

Berry Mingus · Ethics, Policy & Safety · 10 min read

Read the full article on Groundy →

HuggingFace vs GitHub Models vs Replicate: Policy Compliance for Uploaders

Berry Mingus — Tue, 21 Jul 2026 04:27:49 GMT

HuggingFace, GitHub Models, and Replicate enforce independent content policies. Uploaders must verify each hub's rules at release time because acceptance on one platform does.

Berry Mingus · Ethics, Policy & Safety · 15 min read

Read the full article on Groundy →

SAMark Text Watermarking: Paraphrase Robustness and the Policy Gap

Berry Mingus — Mon, 20 Jul 2026 06:30:47 GMT

SAMark reports 90.2% detection under paragraph paraphrase, yet no mandate specifies a robustness threshold. Without a survival metric, disclosure rules reduce to suggestions.

Berry Mingus · Ethics, Policy & Safety · 15 min read

Read the full article on Groundy →

A Digital Twin Can Validate AV Safety, but No Regulator Accepts the Evidence

Berry Mingus — Fri, 17 Jul 2026 05:16:41 GMT

A July 2026 risk-field digital twin replays rare AV hazards no road fleet can accumulate, but no regulator defines when simulated miles count as certification evidence.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Why EU AI Act Monitoring Will Miss Discontinuous LLM Alignment Failures

Berry Mingus — Fri, 17 Jul 2026 03:02:55 GMT

Hair-Trigger Alignment shows an LLM can pass every black-box probe, then flip after one benign update, so EU AI Act periodic monitoring certifies less than it assumes.

Berry Mingus · Ethics, Policy & Safety · 11 min read

Read the full article on Groundy →

Can US Export Controls Contain AI Built Without American Chips?

Berry Mingus — Wed, 08 Jul 2026 13:39:33 GMT

Meituan's unverified LongCat-2.0 claim, a trillion-parameter model on domestic chips, would weaken US chip controls if true. Without proof, the bottleneck question is open.

Berry Mingus · Ethics, Policy & Safety · 8 min read

Read the full article on Groundy →

Entropy Regularization Buys RL Robustness That Certification Can't Credit

Berry Mingus — Wed, 08 Jul 2026 12:42:12 GMT

A July 2026 arXiv preprint proves entropy regularization in continuous-time RL lower-bounds robustness to joint perturbations, but ISO 26262 and IEC 61508 cannot credit it.

Berry Mingus · Ethics, Policy & Safety · 8 min read

Read the full article on Groundy →

Fusion's ML Disruption Predictors Have No Shared Validation Standard

Berry Mingus — Wed, 08 Jul 2026 12:14:07 GMT

A new EAST paper makes real-time tokamak disruption predictors cheap, but ML safety systems lack shared cross-machine validation standards before commercial plants debut.

Berry Mingus · Ethics, Policy & Safety · 8 min read

Read the full article on Groundy →

AI-Generated CSAM Risks Expose Filter-First Safety Gaps

Berry Mingus — Wed, 08 Jul 2026 08:39:41 GMT

A July 2026 ICML spotlight paper argues preventing AI-generated CSAM requires upstream design controls, because auditing, red teaming, and benchmarking cannot include it.

Berry Mingus · Ethics, Policy & Safety · 8 min read

Read the full article on Groundy →