Search: “memory-integrity”

Smart search ranks by meaning as well as keywords (one row per story, last 45 days).

LLMs and Contextual Integrity

Bruce Schneier highlights two papers: the CIMemories benchmark shows frontier LLMs leak memory attributes up to 69%, and an RL method reduces inappropriate disclosures.

Bruce Schneier discusses contextual integrity in LLMs, referencing the CIMemories benchmark, which uses synthetic profiles with 100+ attributes per user to test whether models with persistent memory disclose sensitive information appropriately. Evaluation showed frontier models exhibit up to 69% attribute-level violations, with GPT-5's violation rate rising from 0.1% to 9.6% across 40 tasks and reaching 25.1% with repeated prompting, showing unstable leakage behavior. A second paper introduces a reinforcement learning framework trained on a synthetic 700-example dataset that substantially reduces inappropriate disclosure while maintaining task performance, with improvements transferring to the human-annotated PrivacyLens benchmark.

Schneier on Security·Aug 18, 10:40 UTC · 29d agoAI safety & security

48

MeClear: Cooperative Game-Theoretic Attribution and Risk-Aware Memory Clearance for Long-Horizon LLM Agents

MeClear uses cooperative Shapley attribution to clear harmful memories from long-horizon LLM agents, boosting task recovery by 25.5 points over baselines.

The paper introduces MeClear, a task-conditioned memory clearance framework for long-horizon LLM agents that identifies and selectively suppresses memories with negative downstream utility without permanently altering the persistent memory bank. It combines Leave-One-Out screening with sampled cooperative Shapley attribution to distribute utility across interacting evidence, resolving redundant conflict masking that single-removal evaluations miss. Across ten long dialogue memory pools it achieves 85.9% target recall and 82.3% overall task recovery, a 25.5 percentage-point improvement over LOO baselines.

arXiv cs.AI / cs.LG / cs.CL·Sep 8, 17:46 UTC · 8d agoAI research

32

Forgetting Without Restarting: Execution-State Unlearning for Stateful LLM Agents

Researchers propose provenance-guided selective replay letting LLM agents forget revoked information without restarts, matching full reset behavior.

The paper formalizes execution-state unlearning for stateful LLM agents, requiring that agents behave as if a revoked memory record was never observed across transcripts, compressed memory, tool plans, and KV caches. It proves exact unlearning requires at least T-τ+1 recomputed transitions and that Provenance-Guided Selective Replay attains this bound via a provenance graph, KV cache cropping, and sanitized replay. In audits across three agent suites, nine baselines, and three model families, memory deletion left leakage unchanged, instruction-based forgetting collapsed under elicitation (Leak@probes = 1.00), and selective replay matched full resets at up to 9x fewer recomputed tokens.

arXiv cs.CR·Sep 4, 08:31 UTC · 12d agoAI safety & security1

42

Does Your Agent's Memory Survive a Model Upgrade? A Controlled Study of Memory Portability

A controlled study finds agent memory portability varies sharply: fixed-schema knowledge graphs survive model swaps while compressed notes degrade.

The study compares preserving an agent's history as raw long context, RAG chunks, compressed natural-language notes, or fixed-schema knowledge graphs across model upgrades, using 48 synthetic histories and two open-weight sub-10B-parameter models. Fixed-schema KG accuracy changed by only +0.0004 ± 0.0020 after a writer swap, while compressed NOTES shifted asymmetrically by +9.91 or -13.28 percentage points depending on migration direction. Mixed 50/50 embedding migrations captured only 4.96 of an 11.90-point RAG re-embedding gain; 80% of the NOTES deficit came from information lost at construction, and 81% of the RAG deficit from retrieval failures. Store-only repair of NOTES failed to reach 90% recovery in all 48 cases, while retaining raw histories enabled recovery in 34 of 48 for one direction.

arXiv cs.AI / cs.LG / cs.CL·Sep 4, 16:44 UTC · 12d agoAI research1

32

Revoked but Still Authoritative: An Empirical Study of Revocation Enforcement in Agent-Memory Systems

An empirical study finds no major agent-memory system enforces fact revocation at retrieval, causing agents to act on superseded, unsafe information.

Researchers tested five agent-memory systems across nine policy scenarios, nine models, and six defense conditions, tracking whether revoked facts are returned and acted upon. No system enforces revocation by default: revoked records are returned whenever the revocation label is visible to the retrieval layer, outrank their replacements, and lead agents to unsafe actions. The authors propose a backend-agnostic guard that sits between the agent and any memory store and withholds revoked or conflicting records at retrieval time.

arXiv cs.CR·Sep 8, 05:03 UTC · 8d agoAI safety & security

38

MemSentry: A Framework for Detecting Persistent Memory Poisoning in Agentic AI

MemSentry intercepts persistent-memory writes in agentic AI to catch memory poisoning, reaching 91.7% accuracy with SBERT+LR classification.

Memory poisoning lets adversaries plant crafted content in an agent's long-term memory to suppress security alerts, enable privilege escalation, or override policies without modifying model weights or system prompts. The paper presents MemSentry, a configuration-driven framework that evaluates proposed persistent-memory writes on source trust, semantic risk, attack radius over a dependency DAG, access risk, and a signed security-state delta to issue deterministic Accept, Review, or Quarantine decisions. Across 1,000 GPT-4-generated scenarios on a 20-asset dependency DAG, SBERT+LR achieved 91.7% accuracy and 0.908 macro-F1, all four classifiers detected 100% of external quarantine-class threats, and verified-insider writes are escalated for human review rather than auto-quarantined.

arXiv cs.CR·Sep 8, 13:40 UTC · 8d agoAI safety & security

48

BenchShield: Formal Model-Backed Instrumentation for Reward Integrity in LLM-Agent Evaluation Infrastructure

BenchShield uses lifecycle-model-backed instrumentation to detect reward hacking in LLM-agent benchmarks, lifting full-chain recall to 77-100% at up to 65% lower cost.

The framework grounds reward-hacking detection in a finite lifecycle model of an evaluation's reward-relevant events, combining a static phase-aware taint analysis with runtime infrastructure-side evidence attribution. Evaluation used a human-labeled corpus of 456 adjudicated trajectories drawn from more than 31,000 public agent runs across three benchmarks. BenchShield improves full-chain recall from 23-94% to 77-100% and same-vector coverage from 16-56% to 43-78%, cuts per-task cost by up to 65%, and achieves 96% accuracy detecting reward hacking at runtime.

arXiv cs.CR·Sep 10, 03:10 UTC · 6d agoAI safety & security1

42

Fortunate Recall: Ontology-Driven Memory Lifecycle Management for Persistent Coherence in LLMs

Fortunate Recall introduces ontology-based lifecycle policies for LLM memory, cutting confabulation roughly in half (e.g., 45.1% to 22.4%) versus Mem0.

Fortunate Recall (FR) is a composable policy layer that classifies personal facts into a 10+1 behavioral ontology and applies category-specific lifecycle rules including differential temporal decay, slot-key supersession, event-time validity, and retrieval routing. FR-Bank scores 76.9% on the new 516-question LifecycleBench, ahead of Mem0, A-MEM, Memory-R1, and MemoryOS (61%-70.5%), and 75.2% on LongMemEval-S. End-to-end, confabulation drops from Mem0's 45.1% to 22.4% over answered queries, with the ranking replicating on open-weight Kimi K2.5 and transferring to the independent BEAM benchmark (46.8% vs 32.9%).

arXiv cs.AI / cs.LG / cs.CL·Sep 9, 16:31 UTC · 7d agoAI research

30

CertiFlash: A Formal Verification Framework for Flash Translation Layers in Computational Solid State Drives

CertiFlash provides machine-checked formal verification of SSD flash translation layers, proving isolation, integrity, and ownership invariants to prevent tenant data leaks.

CertiFlash is an open-source formal verification framework for Flash Translation Layers (FTL) in computational SSDs, mechanized in the Rocq proof assistant. It shows that a faulty FTL can corrupt device state at five surfaces (e.g., leaking data between tenants or dropping integrity tags), demonstrated on a DaisyPlus OpenSSD. Designers prove once that every operation of a general FTL model preserves a global invariant covering mapping, isolation, integrity, ownership, and allocation; new designs need only discharge five hypotheses. Across four case studies, added effort was 27-3,231 lines against a 16,489-line framework.

arXiv cs.CR·Sep 9, 15:44 UTC · 7d agoResearch1

25

When Models Edit Too Much: On the Fidelity of Minimal Code Edits

A 400-task BigCodeBench evaluation shows frontier LLMs widely over-edit code; a preservation instruction cuts excess edits and raises Pass@1 by 2.3 points.

Researchers built an evaluation framework from 400 BigCodeBench problems with injected AST-level corruptions, each with a known minimal patch, to measure over-editing in LLM code repair. Even strong models like GPT-5.5 produce unnecessarily large edits despite high Pass@1. Adding a preservation instruction reduced average excess Levenshtein distance from 0.195 to 0.131, cut added cognitive complexity by 26.6%, and raised Pass@1 by 2.3 points. Reinforcement learning post-training gave the best out-of-domain edit-fidelity trade-off, while supervised fine-tuning overfit to seen corruption patterns.

Hugging Face daily papers·Sep 2, 20:00 UTC · 14d agoAI research1

30

Continual Learning Mechanisms Compose for Long-Horizon Memorization

Composing data, function, and weight anchors with merged LoRA raises 100-task long-horizon retention from 1.2% to 34.9% in continual fine-tuning.

The paper introduces long-horizon memorization: a model learns 100 query-answer tasks through continual supervised fine-tuning without retaining earlier examples or receiving task identifiers at inference. No single continual learning mechanism maintains strong retention at this horizon, so the authors compose complementary mechanisms along data/function/weight anchors and low-rank allocation rules. The best method combining all three anchors with merged LoRA ranks among the top 3 methods on all three datasets and raises average final retention from 1.2% to 34.9%, a 28-fold improvement.

Hugging Face daily papers·Sep 6, 20:00 UTC · 10d agoAI research

42

Recognition-Refusal Misalignment in LLMs: Why Models Answer Structurally Unanswerable Questions

A linear hidden-state direction encodes question impossibility in 1.7B-70B LLMs, but misalignment with the safety-refusal pathway explains why models answer unanswerable questions.

The study examines why instruction-tuned LLMs from 1.7B to 70B parameters answer structurally unanswerable math and code questions instead of abstaining. A single linear direction in the hidden state separates answerable from impossible prompts, showing models represent impossibility before generation, but this direction is nearly orthogonal to the canonical safety-refusal direction. Generation-time steering along the recognition direction changes invalidity-aware behavior dose-responsively, and the geometry is present even at the pretraining endpoint, indicating a routing failure rather than an encoding failure.

Hugging Face daily papers·Aug 28, 20:00 UTC · 19d agoAI safety & security

30

CHERI-D Reincarnate: efficient multicore CHERI temporal memory safety through allocation reincarnation (draft version)

CHERI-D Reincarnate adds allocation reincarnation to CHERI, quarantining generation IDs instead of memory to cut temporal memory-safety overheads on multicore systems.

The proposal extends CHERI-D temporal memory safety by quarantining exhausted generation IDs rather than memory slots, allowing freed memory immediate reuse through allocation reincarnation. It adds coherent ID caching and multicore support connecting physical coherence events to the virtually addressed ObjID buffer. The design was implemented as a hardware-software co-design across CHERI-Toooba FPGA softcore, QEMU, LLVM/Clang, and CheriBSD. Across evaluated workloads it substantially reduces memory-sweep frequency and quarantine overhead with low performance and hardware cost.

arXiv cs.CR·Sep 10, 14:19 UTC · 6d agoResearch

27

Bias-Induced Crossover in Absolute Capacity of Dense Associative Memory

Analysis shows biased patterns cut dense associative memory capacity from N^(n-1)/ln N to O(N^(n/2)), with a bias-induced crossover.

The paper analyzes dense associative memory capacity for biased centered binary patterns under the Krotov-Hopfield single-site criterion. Unbiased patterns (q=1/2) with order-n polynomial interactions yield capacity of order N^(n-1)/ln N, while fixed bias q<1/2 reduces capacity to O(N^(n/2)) for even n>=4 and O(N^((n+1)/2)) for odd n>=5. A bias-dependent crosstalk mean destabilizes sites carrying the frequent value, and an activity-dependent control potential restores the higher capacity within the conditioned-Gaussian approximation.

arXiv cs.AI / cs.LG / cs.CL·Sep 15, 17:16 UTC · 1d agoAI research

12

Decomposition Buys Integrity, Not Yield

Study of 600 production deep-research traces finds agent-tree decomposition loses findings at rate N^(1-δ); flat architectures maximize yield.

The paper models multi-agent decomposition as a tree where an agent holding b items retains each with probability r(b); with r(b)=1/b every tree delivers exactly one finding regardless of shape. Analysis of 600 production deep-research traces estimates delta=0.34 retention decay, and 1,012 annotated traces show one brief in sixteen goes off-target per tier, giving an alignment penalty of 0.536. Depth still cuts root context exposure from N to N^(1/k) and is cheaper at scale, with a hazard model over 743,819 production tool calls showing delegation is an opening move rather than a response to filling context.

arXiv cs.AI / cs.LG / cs.CL·Sep 15, 17:04 UTC · 1d agoAI research

30

Testing race conditions with memory access tracing and stack-based delay injection

Google Project Zero released MAccConc, Linux kernel tooling that traces memory accesses to explore and test race condition interleavings.

A Google Project Zero researcher published MAccConc (Memory Access Concurrency), tooling for exploring possible interleavings of multithreaded test cases in the Linux kernel, available on GitHub. The tools use KCOV with ASAN outline-mode instrumentation to record per-access memory traces, enabling automatic testing of all A-B-A interleavings plus terminal and GUI explorers for manual analysis. The work targets confirming race condition candidates, building reliable regression tests, and enabling concurrency fuzzing, drawing on ideas from SKI and Ned Williamson's sockfuzzer.

Google Project Zero·Sep 8, 07:00 UTC · 8d agoResearch1

48

SimpleMemVLA: A Simple but Effective Native-Video Memory for Vision-Language-Action Models

SimpleMemVLA passes full timestamped video history straight to a VLA backbone, setting state of the art on four memory benchmarks.

SimpleMemVLA is a vision-language-action model for long-horizon manipulation that removes the dedicated memory module entirely. It keeps sampled history intact and feeds it to the backbone as timestamped video, with the hidden states of a generated sub-task serving as the only channel into a standard flow-matching action head. Prefilling the shared history prefix during action execution keeps latency close to a single-frame VLA. The system sets a new state of the art on four memory benchmarks and outperforms retrieval, compression and recurrent-state mechanisms, with causal interventions confirming the policy genuinely reads its history.

Hugging Face daily papers·Sep 1, 20:00 UTC · 15d agoAI research

25

Forging Tree-Ring: Reproducing and Instrumenting Black-Box Semantic Watermark Forgery

Reprompt watermark forgery reproduces on Stable Diffusion XL using free-tier T4 GPUs, with forged images accepted by the genuine detector 5 of 6 times.

The authors reproduce the Reprompt forgery attack of Müller et al. against Tree-Ring watermarking on Stable Diffusion XL using the released code on free-tier dual T4 GPUs with 14.6 GB usable memory, versus the A40 hardware of the original study. Over six trials, the genuine detector flagged genuine images 6/6, clean images 0/6, and forged images 5/6, at 325-332 seconds per attack. They also recovered the detector's discarded non-central chi-square statistic and built two natural scores separating forged images from the clean null at AUC 0.861 and 0.972. The notebook, pinned fork, and all measurement artifacts are released with the paper.

arXiv cs.CR·Sep 11, 14:35 UTC · 5d agoResearch

30

Understanding the Impact of Model Pruning on Long-Tail Forgetting and Explanation Reliability in Medical Imaging

Systematic study finds model pruning causes frequency-dependent long-tail forgetting in medical imaging and that gradient-informed methods best preserve explanations.

Across two long-tailed medical imaging datasets, two CNN architectures, four pruning methods, and sparsity up to 95%, the study measures predictive performance, explanation stability, and faithfulness. Rare classes degrade earlier and more severely than frequent ones, while explanation reliability depends mainly on the pruning strategy, with gradient-informed methods degrading least. Mechanistic analysis ties explanation collapse to loss of class-discriminative gradients rather than vanishing feature activations, recommending class- and explanation-aware evaluation of compression.

arXiv cs.AI / cs.LG / cs.CL·Sep 7, 17:44 UTC · 9d agoAI research

27

UniH^3: Unifying Hierarchical Homogeneity and Heterogeneity for All-in-One Medical Image Restoration

UniH3 unifies hierarchical homogeneity and heterogeneity modeling for all-in-one medical image restoration across modalities and degradation types.

UniH3 introduces a Hierarchical Homogeneity Memory module that distills shared anatomical priors from high-quality images, injected via a Homogeneity-Guided Attention mechanism. A Hierarchical Heterogeneity Balancer mitigates inter- and intra-task conflicts during multi-task optimization. It achieves state-of-the-art on MedIR-2D-500K and MedIR-3D-3D benchmarks for both all-in-one and single-task restoration, with code released on GitHub.

Hugging Face daily papers·Sep 9, 20:00 UTC · 7d agoAI research

25

Anatomical Grounding and Leakage-Aware Multimodal Contrastive Learning for Alzheimer's Disease Classification from Structural MRI

Study of label leakage and anatomical grounding in multimodal MRI models for Alzheimer's staging shows cognitive-score fusion accuracy of 87.3% is leakage-driven.

The authors train a ResNet18 slice-based encoder with a one-layer Transformer on 1,075 ADNI-1 T1 MRI scans, using FastSurfer segmentations and YOLOv8 localization (mAP_50 above 0.96) as anatomical reference. Grad-CAM shows the image-only classifier often attends to skull and background rather than disease-relevant structures. A CLIP-style image-tabular contrastive framework organized along a label-leakage spectrum yields 87.3% three-way accuracy with cognitive scores versus 73.0% with regional volumes, and cropping to the medial temporal lobe raises image-only accuracy from 58.7% to 65.1%. Results come from single runs on a small balanced test set with reported confidence intervals.

arXiv cs.AI / cs.LG / cs.CL·Sep 14, 17:13 UTC · 2d agoAI research1

16

When AI Remembers Too Much

Unit 42 PoC shows indirect prompt injection can poison Amazon Bedrock Agent long-term memory, enabling silent exfiltration of conversation history across future sessions.

Palo Alto Networks Unit 42 published a proof of concept showing that indirect prompt injection can silently poison the long-term memory of Amazon Bedrock Agents when the memory feature is enabled. Malicious content on a webpage or document manipulates the agent's session summarization process, so injected instructions persist across sessions and are added to later orchestration prompts, silently exfiltrating user conversation history. The issue is not a vulnerability in the Amazon Bedrock platform but an illustration of the broader unsolved LLM prompt-injection challenge. Amazon reviewed the research and stated that Bedrock Guardrails with the prompt-attack policy provides effective mitigation.

Palo Alto Unit 42·Aug 17, 12:32 UTC · Aug 17, 2026AI safety & security

55

Forgetting Only What Matters: Layer-Selective Unlearning toward Robust LLMs

Researchers introduce FOM-UL, a layer-selective machine unlearning framework that improves forgetting-utility trade-offs and resists knowledge recovery after quantization.

FOM-UL selects transformer layers for unlearning using a forget-to-retain significance score, concentrating parameter updates on layers highly influential for the forget set while leaving most of the model unchanged. It reduces residual memorization versus GA, NPO, KLD, SURE, ReLearn, and LUNAR-based baselines on TOFU, KnowUnDo, and MUSE-style evaluations while preserving retain-set utility. Under 8-bit and 4-bit post-training quantization and adversarial prompts, it maintains stronger suppression of forgotten content, addressing brittleness of diffuse unlearning updates.

arXiv cs.AI / cs.LG / cs.CL·Sep 9, 16:50 UTC · 7d agoAI research1

36

I accidentally turned LLM memory into program analysis

A pwning.systems write-up describes how LLM memory functionality was unexpectedly repurposed into a program analysis technique.

A security research post on pwning.systems describes the author's discovery that LLM memory behavior effectively functioned as program analysis. The write-up is hosted on a security-focused blog and surfaced via a security-tagged link aggregator. Detailed technical content is not included in this feed, limiting verifiable specifics.

Lobsters · security·Aug 29, 03:16 UTC · 18d agoResearch1

30

Towards Standardized Evaluation of GPU Memory Safety with GMSBench

GMSBench provides 149 CUDA tests covering spatial, temporal, and concurrency GPU memory errors, exposing detection gaps in Compute Sanitizer.

GMSBench is a GPU memory safety benchmark comprising 149 self-contained CUDA tests spanning spatial, temporal, and concurrency errors across different GPU memory spaces and execution scenarios. The authors evaluate NVIDIA's Compute Sanitizer across multiple GPU architectures using the suite, exposing gaps in its detection coverage. The benchmark offers a standardized foundation for comparative evaluation of GPU memory safety mechanisms.

arXiv cs.CR·Sep 8, 15:11 UTC · 8d agoResearch

28

TasmScan: Continuation-Aware Taint Analysis for TVM Bytecode with Savelist Abstraction

TasmScan introduces source-free taint analysis for TON smart-contract bytecode, detecting 95.3% of defects with 96.8% precision and 17x speedup.

TasmScan is the first bytecode-level static analysis framework for the TON Virtual Machine, enabling cross-continuation data flow reasoning without source code by modeling savelist semantics through forward register analysis with formal over-approximation guarantees. It lifts bytecode into a typed intermediate representation (TASIR) and performs path-sensitive taint analysis. On a 208-contract benchmark with human-confirmed ground truth it detects 95.3% of defects across five classes at 96.8% precision, and resolves 294,546 dynamic continuation targets with 100% precision across 2,921 registry contracts. It achieves a 17x median speedup over symbolic-execution baselines.

arXiv cs.CR·Sep 15, 11:01 UTC · 1d agoResearch

28

A Chosen Future Can Still Be Rewritten: Causal Writability in Video Models

Study shows video models often learn correct physics but fail to use it; low-dimensional 'causal writability' edits can restore correct motion.

The paper demonstrates 'causal writability' in video generation models: physically correct motion remains available inside the model even when the model outputs incorrect motion. In a red/blue mass oscillation setup, a low-dimensional edit predicted from simple physical variables restores correct fast motion, with a sharp depth boundary marking commitment. Early causal writability predicts which training errors later get corrected, and both writability and closure reproduce in a pretrained 1.3B video model.

arXiv cs.AI / cs.LG / cs.CL·Sep 14, 17:58 UTC · 2d agoAI research

33

Reference-Based Bias Detection in LLMs via Relative Representations of Hidden States

Researchers propose auditing LLM bias via relative hidden-state representations, detecting bias increases with 3-50x less compute than output-level benchmarks.

The paper introduces a reference-based bias auditing method that compares hidden-state representations across model variants, such as before and after fine-tuning, by encoding sentences relative to a fixed anchor set. The resulting Representational Bias Shift (Delta-B) correlates with output-level bias change in 15 of 18 tested settings, reaching |r| = 0.84 under full fine-tuning across WildGuardMix, DecodingTrust, and ToxiGen benchmarks. Thresholding Delta-B detects checkpoints whose bias increased with ROC AUC between 0.65 and 0.99 and beats a SEAT-based baseline, while auditing a model in about three minutes with 3-50x less compute.

Hugging Face daily papers·Sep 8, 20:00 UTC · 8d agoAI research1

42

Sound Debloating of Redundant Checks in Zero-Knowledge Machine-Learning Circuits

Automated framework soundly removes up to 48.7% of redundant constraints in ezkl and zkml ZK-ML circuits, cutting prover time by up to 72.8%.

The framework uses whole-circuit abstract interpretation and a provenance graph to verify that each removed redundant check (range proofs, sign lookups, bit decompositions) remains entailed by the rest of the circuit, provably preserving soundness. It was evaluated on MLP, CNN, RNN, and transformer circuits generated by ezkl and zkml, with up to 25.3 million constraints. It removes up to 48.7% of constraints and reduces prover time by up to 72.8% without weakening security. Under-constrained circuits in deployed ZK systems have previously enabled attackers to forge transactions and bypass verification.

arXiv cs.CR·Sep 9, 13:27 UTC · 7d agoResearch1

35

Beyond Solver Verdicts: Generative Reward Models for Autoformalization

Researchers introduce Generative Verification (GenV), a generative reward model achieving 0.961 AUROC in detecting unfaithful autoformalization that preserves solver verdicts.

The paper formalizes Verdict-Preserving-Unfaithfulness (VPU), a failure mode in neurosymbolic autoformalization where an incorrect encoding executes successfully and matches the expected solver verdict, and proves verdict-only verification is bounded to chance-level detection. The proposed Generative Verification (GenV) distills an offline Z3-equivalence oracle into a reference-free, continuous reference-equivalence score within the language model's vocabulary space. The oracle-mined verifier (GenV+HN) achieves 0.961 AUROC, generalizes zero-shot across unseen translators and formal styles, and yields an 11.3-point downstream accuracy gain in agentic test-time compute allocation. Mechanistic analysis with decision-projected logit lenses and sparse autoencoders shows the generative readout extracts precise spatial error coordinates without explicit localization training.

Hugging Face daily papers·Sep 9, 20:00 UTC · 7d agoAI research1

35