AI News — 2026-10 · Enkuklios

This page is the month-by-month index of the AI Pedia’s Recent News. Dated news articles are grouped here instead of being listed inline in the hub, so news.md and the sub-article index stay concise. Each article is listed with its one-line summary.

2026-10-28

An Algebra of Transformers — Gives transformers a logical algebra for reasoning and verification
Artificial Analysis Intelligence Index v5.0 — Artificial Analysis released v5.0 of its Intelligence Index, a widely cited cross-model benchmarking methodology that ranks frontier models
Benchmark Radar: A Living Database and Search Engine for AI Benchmarks and Evaluation — Living database and search engine for AI benchmarks and evaluation
CalArena: A Large-Scale Post-Hoc Calibration Benchmark — Reliable probability estimates are critical in many machine learning applications, yet modern classifiers are often poorly calibrated
Can LLMs Reliably Grade Olympiad Proofs? A Study of Mathematical Verification with LLMs — A systematic study of LLM-based verification of Olympiad-level proofs
Context Is a Pipeline, Not a Window: Information Attrition in LLM Agent Harnesses — Measured tool-output loss in agent harnesses costs extra work even when answers stay correct
Cross-User Poisoning: User-Task Boundary Failures in Multi-User Collaborative Language Agents — Language agents are moving from single-user assistants to shared collaborators in workspaces, forums, and group chats
Detecting and Mitigating Query-Based System Prompt Leakage Attacks — AdvGuard detects coordinated multi-turn attempts to extract system prompts
Fairness Definitions in Language Models Explained — Language Models (LMs) have demonstrated exceptional performance across various Natural Language Processing (NLP) tasks
Follow the Gradient: Tracing Safety Degradation Back to Training Data — Traces fine-tuning safety erosion back to specific training data
From SGD to Muon: Adaptive Optimization via Schatten-p Norms — Generalises matrix-wise adaptive optimization through Schatten-p norms
HOLogicBench: Evaluating Higher-Order Logical Reasoning in LLMs — Benchmark for higher-order logical reasoning in LLMs
How K-12 Teachers Adapt Generative AI to Local Classroom Demands: Cultural Distance Between GenAI Defaults and Everyday Teaching Practices — Generative AI (GenAI) is entering K-12 classrooms, but its default culture does not always fit local teaching. Existing accounts either trea
Log-To-Leak: Prompt Injection Attacks on Tool-Using LLM Agents via the Model Context Protocol — Additive post-hoc logging attack leaks data while preserving legitimate task completion
Moderation by Interrogation: Uncovering Concealed Intent in Agent Social Networks — Probing agent networks to uncover concealed intent
MuonGS: Pretraining LLMs with Relaxed Weight-Norm Control — Relaxed weight-norm control for LLM pretraining
PIVOT: Bridging Planning and Execution in LLM Agents via Trajectory Refinement — Refines agent trajectories to bridge planning and execution
Preconditioned Flow Matching — Flow matching (FM) learns vector fields by regressing stochastic velocity targets along intermediate distributions $p_t$
ScalableRAG: High-Quality RAG with Exact Corpus Operations at Minimal Ingestion Cost — High-quality RAG with exact corpus operations and minimal ingestion
Scaling and Renormalization in Neural Networks: An Effective Renormalization-Group Description of Grokking — Real-space RG gives a quantitative description of grokking transitions
Semantic World Models — Planning with world models offers a powerful paradigm for robotic control. Conventional approaches train a model to predict future frames co
Specialization, Diversity, and Scaling in Child-Scale Language Models — Small models trained on child-scale input specialize to their environment; diversity, not just volume, shapes grammar and semantics
Suan: Rectifying Direct Preference Safety Alignment in Large Language Models — Integrating robust safety guardrails into Large Language Models (LLMs) is essential for delivering helpful yet harmless responses
TextCloak: Safeguarding Data from Unauthorized LLM Fine-Tuning via RL-Driven Unlearnable Examples — The development of Large Language Models (LLMs) has led to significant advances across a wide range of language tasks, while simultaneously
The Fake Friend Dilemma: Relational Trust and the Political Economy of Conversational AI — As conversational AI systems become a larger part of the media landscape, they raise questions about whose interests they serve and the risk
The Structure Gap: When More Data Cannot Replace Structured Reasoning — Argues data alone cannot replace structured reasoning
The Trinity:Capability-Driven Recursive Self-Improvement for Multimodal Agents — Multimodal intelligence can be viewed as human-guided co-evolution between worlds and agents through interaction. However, scaling this process faces two bottlenecks: constructing heterogeneous, verifiable…
The uneven rise of AI-generated web content — AI-generated web content has risen precipitously but unevenly across languages and domains
ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning — World imagination reinforces chain-of-dynamics reasoning for robots
Towards Human-AI Complementarity for AI Safety — Safely overseeing AI is increasingly challenging for the humans doing it, as they must evaluate longer and more complex outputs with every n
Understanding the limits of latent world models — Studies the limits of latent world models for agents
What Makes Video World Model Latents Action-Relevant: Prediction over Reconstruction — Video world models are increasingly used to provide predictive visual representations, yet it remains unclear which pretraining signals indu
When AI Agrees Too Much: Manipulating Automated Code Review via Bias Injection — Automated Code Review (ACR) systems integrating Large Language Models (LLMs) are increasingly adopted in software development workflows, ran
When to Intervene? An Empirical Study of Errors and Self-Recovery in LLM Agents — Empirical study of errors and self-recovery in LLM agents
Why and How to Do Science When AI Is Better at It: A Framework Grounded in Human Understanding and Direction — As AI begins to match or exceed human researchers at core scientific tasks, two questions become pressing: \emph{why} should humans keep doing science, and \emph{how} should they use AI when they do? This paper…

2026-10-27

A Comprehensive Framework for All Stages of Large-Model Training: From Data and Tokenization, through Pretraining, to Post-Training Alignment — The capabilities of large language models (LLMs) and multimodal LLMs (MLLMs) arise from the synergy of the full training pipeline, including data and tokenizati
A Deterministic Security Runtime for LLM Agents on MPI and Batch-Scheduled HPC Systems: Consensus-Governed, Hardware-Rooted Supervision of Every Rank — Large language model agents launch simulations on high-performance computing (HPC) systems, where one sbatch or mpirun can commit thousands of node-hours and a
A Distribution-Matching Perspective on LLM Training Algorithms — In this work, we propose a unified framework for training algorithms across different stages of modern large language model development, based on the perspectiv
A False Average: Pooled CoT-Monitor Accuracy Conceals a Reasoning-Dependent Fragility — Pooled CoT-monitor accuracy conceals a reasoning-dependent fragility
A Survey on Long-Term Memory Security in LLM Agents: Attacks, Defenses, and Governance Across the Memory Lifecycle — The emergence of writable, cross-session persistent memory in LLM agents introduces a qualitatively different threat landscape from conventional input-centric security concerns, characterized by three properties:…
Agent Memory Is a Surface for Endogenous Authorization Laundering — Memory misrepresentation can launder agent authorization
Agentic Societies Need a Social Harness — Norms and protocols needed for multi-agent coordination
Agents’ Last Exam — Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional d…
Application of Artificial Intelligence in the Management of Climate Change-Related Health Issues in Africa — While climate change disproportionately impacts African public health, AI predictive modeling remains heavily constrained
Babysitting AI Agents: A User-Centered Taxonomy and Study of Agentic Delegation Failures — Users are increasingly delegating tasks to AI agents, allowing agents to perform long-running, multi-step workflows on their behalf. However, agents may perform these tasks in ways…
Beyond Behaviour: A Stimulus-Response Safety Evaluation of Vision-Language-Action Models — Humans exhibit biological sensitivity to safety-critical stimuli that can trigger defensive action
Beyond Belief: Evaluating and Mitigating LLM Susceptibility to Misinformation — Tool-using large language models (LLMs) must decide whether external information warrants action
Beyond Reasoning Gains: Mitigating General-Capability Forgetting in Large Reasoning Models — Reinforcement learning with verifiable rewards (RLVR) has delivered impressive gains in mathematical and multimodal reasoning and has become a standard post-training paradigm for c…
Beyond Single-Attribute Auditing: A Configurable Multi-Attribute Benchmark for Intersectional Fairness in LLMs — Large language models (LLMs) are increasingly evaluated for fairness, yet most existing benchmarks audit one sensitive attribute at a time
Broken Symmetry in LLM Refusal: Answer Release Is More Local Than Refusal Restoration — When a language model refuses to answer a prompt, it is unclear whether the correct answer is erased from its internal representations, or merely suppressed at the output layer. We…
Build, Don’t Play: Self-Improving LLM Research Agents Develop a Hearthstone Battlegrounds AI That Rivals Top Players — Can LLM agents carry a hard real-world problem from raw data to a system that holds its own against people? We study this in Hearthstone Battlegrounds, an eight
C-SafeGen: Certified Safe LLM Generation with Claim-Based Streaming Guardrails — Despite the remarkable capabilities of large language models (LLMs) across diverse applications, they remain vulnerable to generating content that violates safe
Can Agentic Code Review Catch Vulnerabilities Before They Are Merged? — LLM agents have been increasingly adopted for security-focused code review, but existing benchmarks mostly derive the reviewed changes from their later fixes in
ChatGPT Images 2.5 in the Wild: A Launch-Period Dataset and Detector Evaluation — Detector evaluation across a ChatGPT Images 2.5 launch period
ClawMMU: Protection-Coherent Virtual Memory for Stateful LLM Agents — Virtual-memory-style isolation for stateful agent memory
Cohere says Alibaba’s Qwen model tried to send a file to Beijing
Compaction as Epistemic Failure: How Agentic LLM Tools Fabricate Confirmed Results from Killed Processes — Compaction as epistemic failure: agentic LLM tools fabricate confirmed results from killed processes
Conformance as a Distribution: Delegation Contracts, Error Budgets, and the Allocation of Evaluation Authority in Use-Based AI Regulation — Use-based AI regulation assigns obligations by deployment risk but lacks an evidentiary object to carry them
Continuous Diffusion Scales Competitively with Discrete Diffusion for Language — While diffusion has drawn considerable recent attention from the language modeling community, continuous diffusion has appeared less scalable than discrete approaches. To challenge this belief we revisit Plaid, a…
Continuum Attention — Continuous piecewise-linear memory trajectory for causal attention
CoSA: The Conflict of Skills Attack in Real-World LLM Agent Ecosystems — Agent Skills allow LLM agents to combine reusable capabilities, but checking each Skill in isolation cannot rule out policy violations when they are composed. We term such failures…
CTXAgent - Self-Controlled Context Management of LLM Agents — Agent harnesses keep long sessions inside the context window by compacting the history when a token threshold is crossed
CyberDual: AI Attackers versus AI Code Reviewers in Real Software Maintenance — Open-source projects now let AI agents review incoming code, but the code under review may itself come from an AI attacker
Data Governance for Generative and Agentic AI : Technology, Law, Economics, and Security — This book develops a unified governance framework for systems that retrieve, generate, infer, remember and act on information from large language models and ret
Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents — Runtime guard skill against malicious skill-augmented agents
Developer Task Delegation to LLMs and the Trust Factor: A Mixed Methods Study — With the rise of Large Language Models (LLMs), software developers are presented with a powerful new approach: delegating software development tasks to LLMs
Diagnosing and Mitigating Context Rot in Long-horizon Search — Diagnosing and mitigating context rot in long-horizon search
Distilling Expert Judgment at Scale — Frontier AI models are increasingly deployed on topics where the stakes are high and expert judgment is required, including healthcare, defense, finance, govern
Do Latent Tokens Think? A Causal and Adversarial Analysis of Chain-of-Continuous-Thought — Latent tokens are gaining attention for enhancing reasoning in large language models (LLMs), yet their internal mechanisms remain unclear. This paper examines the problem from a re…
Early progress on accelerating drug development with Claude — Anthropic reports early progress using Claude to accelerate drug development, applying frontier models to scientific discovery workflows. It
Evaluating Harmful Memorization in LLM Code Generation — Assessing memorization in LLM code generation is challenging because code reuse is not inherently harmful. Correct programs often share APIs, templates, and algorithmic idioms, so …
Evaluating LLM-based Agents for Multi-Turn Conversations: A Survey — This survey examines evaluation methods for large language model (LLM)-based agents in multi-turn conversational settings. Using a PRISMA-inspired framework, we systematically revi…
Evaluating Visual Language Models on Visual Grounding for Visually Impaired People — Visual grounding (VG) in visual question answering (VQA) is a fundamental capability of Vision-Language Models (VLMs), enabling the detection of the most releva
Fantastic Pretraining Optimizers and Where to Find Them — AdamW has long been the dominant optimizer in language model pretraining, despite numerous claims that alternative optimizers offer 1.4 to 2x speedup. We posit that two methodologi…
From AI-Free Gates to AI-First Work: How Engineering Students in India Navigate University Education and Career Preparation Without AI Guidelines — Indian engineering students navigate AI-free exams into an AI-first workforce
Giving chat feedback gives Anthropic your data — Claude’s feedback controls submit conversation data for training
Google’s Genie 4 world model adds audio, more realistic worlds, and real-time steering — Genie 4 signals that world models are moving beyond static image or video prediction toward interactive, multi-sensory environments a user can steer in real time. Such models could underpin game-like experiences,…
Harness Scaling for Generalizable Agentic Reinforcement Learning — Scaling agent harnesses rather than only models improves agentic RL
Harness-Model Collaboration for Long-Horizon Continual Learning — Long-horizon agents continuously encounter reusable procedural knowledge, yet existing systems typically rely either on model adaptation or external memory, tra
Here Be Monsters: What Do You Still Need to Know When Agents Do the Work? — Essay on human oversight as agents take over more tasks
How Do Coding-Agent Harnesses Self-Evolve? An Empirical Study of Retrospective Adaptation — Coding agents can adapt their harnesses, the software, configuration, and instructions used to run them, based on experience from earlier tasks, but whether these adaptations impro…
How Firm Are AI Models’ Ethical Boundaries? A Pay-Stub Editing Test of 15 Models — Can a small change in wording move an AI model from refusing a sensitive request to acting on it? We test 15 models on the same request to increase the income shown on a pay stub. …
How Much Information Can We Fit inside a Transformer? — How efficiently can transformers store factual knowledge? The answer to this question has implications for both model design and understanding the limits of tra
How OpenAI turned its coding tools into dots, an always-on personal assistant
How To Build A Modern Diffusion Language Model — The frontier of large language model (LLM) performance has advanced rapidly in recent years, fueled by scaling laws for pretraining on large text corpora
HyScale: Scaling Long-Context Training for Learned Sparse Attention — We present HyScale, a long-context training system for learned sparse attention. Learned sparse attention is a promising architecture for million-token contexts, widely adopted in …
Identically Trained Language Models Break on Different Jailbreaks — Jailbreak benchmarks test one checkpoint and treat which jailbreaks break it as a property of its pretraining data and safety training
Infinity Machine: a self-improving AI agent for research — We present the Infinity Machine, a self-improving AI agent that we use for research on problems whose answers aprogram or data can check
LLM WikiLeaks: Uncovering URL-Based Capability Extension and Security Circumvention in Agentic Message Boards — Autonomous agents have been observed acting beyond their assigned tasks, accessing unauthorized systems, and establishing communication channels to exchange inf
LLMs are not stochastic parrots: Evidence for meaning-mediated abstraction from conlang-like tasks — Evidence for meaning-mediated abstraction challenging the stochastic-parrot claim
Measurement Without Validity: The Compounding Reliability Problem in Agentic AI Evaluation — Agentic AI evaluation pipelines produce benchmark scores that justify deployment decisions, safety certifications, and regulatory compliance claims. No formal framework has charact…
NameTrap: Steering AI Coding Agents Toward Security Vulnerabilities via Variable Names — AI coding agents routinely clone repositories, install packages, and read source code obtained from the Web, so publicly hosted code becomes part of the context
NewsBench: Expert-Grounded Evaluation of Epistemic Quality in AI News Reporting — As frontier AI systems increasingly serve as sources of political and current-events information, evaluating the epistemic quality of their responses is now esp
Nvidia backs Indian startup helping companies run AI models with fewer GPUs
One Leak Away: How Pretrained Model Exposure Amplifies Jailbreak Risks in Finetuned LLMs — Finetuning pretrained large language models (LLMs) has become the standard paradigm for developing downstream applications. However, its security implications remain unclear, parti…
OS-Wizard: Scaling Computer-Using Agents to Specialized Software via Knowledge-Guided Exploration — Computer-using agents are increasingly capable of performing complex tasks in the digital world
Parents sue OpenAI, alleging GPT-4o coached their daughter off seizure medication — This lawsuit underscores the real-world stakes of giving consumers unrestricted conversational AI for health questions. It raises questions about model design, disclosures, and liability when a general assistant’s…
Profiling a Frontier Agent with EBench: From Direct Control to Harness with Policy — Frontier multimodal large language models (MLLMs) can control robots without robot-specific training, but how their capabilities compare and interact with post-trained vision-langu…
Propensities Please: Exposing Alignment-Faking LLMs using Synthetic Document Finetuning — Alignment evaluations aim to assess whether large language models would engage in dangerous misaligned behaviour
Recoverability-Controlled Safety Miscalibration in Large Language Models — Separating intent understanding from response-level safety in evaluations
Reva: Real-Time Vulnerability Detection through Contextual Probing of Large Language Models — A vulnerability costs more to fix the later it is discovered. As coding agents generate code faster than developers can review it, software needs a first line of defense screening …
Roblox downloaded 145GB of models from Open3DLab without asking — Large-scale scraping of 3D model repositories raises data-provenance questions
RSI Arena: Towards Recursive Self-Improvement with Human Feedback — Can AI agents autonomously train useful language models and improve them using feedback from users? We introduce RSI Arena, a two-stage study of autonomous post
Safe by Benchmark, Unsafe in Practice: Lessons from 1331 Child-Related AI Incidents — AI systems are increasingly present in children’s everyday environments: education, entertainment, social networks, and toys
Safeguard-Conditioned Uplift: Measuring Utility-Risk Frontiers for Dual-Use Biology Assistants — Safeguard-conditioned uplift measures utility-risk frontiers for dual-use biology assistants
Scaling Laws and Tradeoffs in Recurrent Networks of Expressive Neurons — Cortical neurons are complex, multi-timescale processors wired into recurrent circuits, shaped by long evolutionary pressure under stringent biological constraints. Mainstream machine learning, by contrast,…
Scaling Verifiable Environments for Long-horizon Work Agents — Scaling verifiable training environments for work agents
Scope-Gap Injection: Exploiting Evaluator Channel Blind Spots in the OpenClaw Agent Security Competition — This paper introduces Scope-Gap Injection (SGI), an indirect prompt injection method for multi-stage LLM-agent safety pipelines
ShutdownMod: Turning Any Trajectory into a Shutdown Evaluation — Theory and experiments suggest that capable AI agents may resist shutdown
Stealth Apart, Harm Together: Skill Cascading Attacks on Skill-Based Agent Systems — A skill is a modular package of natural-language instructions, executable scripts, and reference resources that an agent can load at runtime to extend its capabilities for a specif…
Terminal-RSI: Recursive Self-Improving Terminal Agents via Recursive Self-Rewrite and Policy-Refine — Converting verifier outcomes into useful policy updates and partial progress into complete task resolution is crucial for terminal agents. We present Terminal-RSI, which alternates…
The Watchtower Imperative: Lifespan-Calibrated Evaluation for Frontier AI — The gap between AI capability deployment and safety assurance is structural and cannot be closed through internal evaluations alone. Analyzing documented harms (2017-2026) across t…
Training an MoE LLM in MXFP4: How We Closed the Gap to BF16 — Training large language models (LLMs) in sub-8-bit arithmetic promises substantial gains in memory bandwidth and compute throughput, yet FP4 formats introduce quantization error se…
Trustworthy Recursive Self-Improvement in the Era of Large Models: Foundation, Risks, Evaluation and Open Challenges — AI systems built around large models increasingly use experience and feedback to make persistent changes to models, prompts, memories, tools, and workflows
Video World Models Need Predictable Latents — Video world models trained on large-scale robot data promise a safe and scalable approach for learning and evaluating policies that act in the physical world. However, they still s…
Vulnerability Analysis for Safe Multi-Agent Reinforcement Learning — Safe Multi-Agent Reinforcement Learning (SMARL) aims to learn cooperative control policies under prescribed safety constraints. Al- though existing SMARL methods focus on enforcing…
Web Agents Should Adopt the Compile-and-Execute Paradigm — ReAct has become the default architecture across LLM agents, and many existing web agents follow this paradigm. We argue that it is the wrong default for web agents. Instead, web agents should default to…
What if automating AI R&D triggers an intelligence explosion? — Whether automating AI R&D could accelerate an intelligence explosion
What Open-Source Projects Tell Their Coding Agents: A Census of AGENTS.md and CLAUDE.md — Census of 22,854 repositories maps how projects guide coding agents

2026-10-26

Anthropic officially files for its public debut
Cross-request KV cache exposure in the GLM-5.3 Workers AI model — Cloudflare discloses a cache-isolation leak affecting 155 responses
The Exponential Time Hypothesis Is False — OpenAI preprint challenges a foundational conjecture in computational complexity
Three new discoveries in mathematics, computer science, and physics — OpenAI announces scientific results, including a preprint claiming the Exponential Time Hypothesis is false

Sources

AI Pedia news pipeline backlog: https://enkuklios.com/backlog.json
Recent News · Sub-Article Index