55 results for “Oj”

StripWake is a projector-only backdoor attack on large vision-language models. It stays dormant under full-token inference and activates when visual-token compression prunes the context around a trigger (context stripping). On LLaVA-1.5-7B across nine unseen compressors, it reaches 0.4% full-token attack success rate and 94.0%-99.7% compressed ASR on five primary compressors, with no CIDEr loss. Training aligns early visual attention with a frozen benign teacher.

arxiv.orgpaperAdded 3 Nov 2026, published 31 Oct 2026🛡️ Security

JADE (Joint Action DEnoiser) is a diffusion-based vision-language-action architecture in which the VLM reads the current noisy action latents together with images and instructions at every denoising step, and guides the action expert through gated cross-attention. The VLM is also trained to recover clean action latents, and an adaptive step-size schedule cuts sampling steps. On RoboDojo it averages 12.67% success, the highest among π0.5-based methods, and it adds 20.8 points over π0.5 on four real bimanual tasks.

arxiv.orgpaperAdded 3 Nov 2026, published 31 Oct 2026🧠 AI

ShiftRAG is a decoupled retrieval-augmented generation framework that uses one LLM backbone but separates discriminative retrieval learning from autoregressive generation. It projects retrieval alignments into continuous soft tokens so the generator receives retriever-side signals, and adds a soft orthogonality objective against rank collapse in decoder-only models. The abstract reports gains in retrieval and question answering over state-of-the-art baselines, without numbers.

arxiv.orgpaperAdded 3 Nov 2026, published 31 Oct 2026🧠 AI

AgentCap is a runtime authorization layer for LLM agents that moves the authorize decision out of the model and mediates every tool call through an MCP wrapper. It issues short-lived, budget-bounded capability tokens checked at the tool boundary, restricts delegation to attenuating parent-child chains, and uses a declarative policy language. On AgentDojo and InjecAgent it lowers attack success in every tested setting, reaching 0% with GPT-5.4.

arxiv.orgpaperAdded 3 Nov 2026, published 31 Oct 2026🛡️ Security

AgentCap is a runtime authorization layer for LLM agents from Virginia Tech (Suresh and Gao), posted to arXiv on 31 October 2026. The model proposes MCP tool calls, and a policy engine issues Ed25519-signed capability tokens only after checks on provenance, intent, trust and prerequisites. It reports 0% attack success with GPT-5.4 on AgentDojo and InjecAgent, but only on scenarios the authors built.

arxiv.orgpaperAdded 3 Nov 2026, published 31 Oct 2026🧠 AI

SecPO is a prompt injection defense that combines a new preference-optimization loss with adversarial training. It replaces the insecure DPO reference model with an approximated injection-invariant oracle, and trains on strong injections synthesized at best@240. Against Meta-SecAlign, AT cuts attack success rates 1.5-14.5x, and on Qwen3.6-27B AgentDojo ASR@800 falls to 8.8% from 78.8%, with no noticeable utility loss across nine benchmarks. Code, training data and model are released.

arxiv.orgpaperAdded 3 Nov 2026, published 30 Oct 2026🛡️ Security

SecPO is a DPO-variant loss from UC Berkeley for training LLMs against indirect prompt injection. It replaces the undefended-model reference on injected input with its output on the clean input, a one-line change, and pairs with a PAIR-based adversarial training pipeline. On Qwen3.6-27B, AgentDojo ASR@800 is 8.8% against 78.8% for Meta-SecAlign, with utility unchanged across nine benchmarks. Code, training set and weights are released.

arxiv.orgpaperAdded 3 Nov 2026, published 30 Oct 2026🧠 AI

SEAL is a defense against indirect prompt injection for local LLM agents. It learns a soft prompt and scoring readout offline, then uses a frozen task model to score and delete injected tokens from tool outputs, with queries built from trusted initial context. On AgentDojo, attack success falls from 34.6% to 0.1% and Joint rises from 51.4% to 77.5%, using about 127x fewer defense tokens than PISanitizer.

arxiv.orgpaperAdded 31 Oct 2026, published 29 Oct 2026🛡️ Security

The paper folds the per-key logit bias of Attention Matching (AM) KV cache compaction into the keys along one closed-form direction per head, fitted on context queries. The result is a plain cache that FlashAttention and vLLM paged caches can use, with output error bounded by how far each query's projection on that direction is from 1. Unmodified vLLM serves folded caches through a KV connector. On QuALITY and LongHealth the fold is within noise of AM in 13 of 14 settings. Under AM's published per-head budgets it is within noise in 9 of 10.

arxiv.orgpaperAdded 28 Oct 2026, published 27 Oct 2026🧠 AI

SymLift is a compile-free vulnerability detection framework for code fragments that lack project context. A fine-tuned LLM reconstructs control-flow, data-flow, memory and resource relations as a constrained LLVM-like IR, and a deterministic parser turns it into facts for Datalog rules that give findings with derivation traces. SymLift-Qwen-7B reaches 94.41% Target-F1 on FormAI and 91.66% micro-averaged Target-F1 on MegaVul.

arxiv.orgpaperAdded 28 Oct 2026, published 25 Oct 2026🧠 AI

OWLS (Open-world Language-conditioned Speed Fields) plans directly over a vision-language model's dense feature map. A text prompt describing traversable terrain scores every pixel of one camera image. The scores are projected onto the ground, adapted with two morphological operations, and used as the speed field of a minimum-time planner. It needs no training, terrain labels, map or extra sensors. It was tested on a physical rover, on simulated Mars and Moon rovers over 200 procedural terrains, and on real planetary imagery.

arxiv.orgpaperAdded 26 Oct 2026, published 23 Oct 2026🧠 AI

Paper on memory injection, where prompt injection persists across sessions through the long-term memory of LLM-as-a-service platforms. It presents a general threat model and shows all NIST attack goals are achievable on ChatGPT Instant 5.3, Claude Sonnet 4.6 and Gemini 3 Flash. Vectors include personal memories, projects, skills, organization instructions and MCP resources. On Claude, shared memory lets an attacker compromise multiple users in an organization and exfiltrate prompts. Nearly half of the derived mitigations are unimplemented.

arxiv.orgpaperAdded 24 Oct 2026, published 20 Oct 2026🛡️ Security

Membership Inference in Privacy-Preserving Vector Search is an IACR ePrint paper (2026/2570) that attacks backend-compatible protections for embedding databases using one snapshot and known plaintexts. It breaks the Scale-and-Perturb scheme of Fuchsbauer et al. (advantage 0.997 on MS-MARCO) and analyzes Rotate-and-Perturb, ideal DCRE and an encrypted HNSW index. It also tests shuffling and secret projection as mitigations.

eprint.iacr.orgpaperAdded 21 Oct 2026, published 20 Oct 2026🛡️ Security

RAISE (Risk-Aware Intensive Supplemental Execution Framework) is a just-in-time framework that predicts which project-days introduce a vulnerability, to decide where to spend extra fuzzing. The dataset joins OSV vulnerability-introducing commits with OSS-Fuzz coverage data for 15 C/C++ projects (15,210 project-days, 347 positive). A Random Forest using VCCFinder static features plus fuzzing features reaches ROC-AUC 0.74. In a simulation over 968 vulnerabilities, Simple-Regression needed 91.60 extra fuzzing sessions per detected vulnerability, against 171.89 for the Median heuristic.

arxiv.orgpaperAdded 22 Oct 2026, published 20 Oct 2026🛡️ Security

Predictable Scaling (PS) is a framework that predicts full-scale recommendation model performance from scaled-down proxies. It builds proxies with composable scaling functions along separate resource axes, then projects metrics to target scale by curve fitting with calibrated uncertainty. Of three predictors, even the raw proxy falls within repeated-run noise, and a Gaussian Process gives 15% lower relative MAE. On a task-specific ranker and a foundation ranker, PS ranks hypotheses correctly at 79-88% compute reduction.

arxiv.orgpaperAdded 20 Oct 2026, published 16 Oct 2026🧠 AI

Ignored, Not Refused audits 33,000 trajectories from eleven open-weight agents (2-14B parameters) on InjecAgent and AgentDojo. A label-blind LLM judge finds most failed indirect prompt injections are ignored (76-80%) rather than refused (17-18%). Changing only the system prompt moves InjecAgent ASR from 0.2% to 43%. The Attack-Task Outcome Matrix (ATOM) shows 21.7% of AgentDojo episodes were safe but left the user task incomplete. The corpus, judge prompt and code are released.

arxiv.orgpaperAdded 20 Oct 2026, published 16 Oct 2026🛡️ Security

"Ignored, Not Refused" audits 33,000 trajectories from eleven open-weight agents (2-14B parameters) on InjecAgent and AgentDojo. A label-blind LLM judge finds 76-80% of failed injection attacks were ignored and 17-18% refused. Changing only the system prompt moves InjecAgent attack success rate from 0.2% to 43%, and the ATOM matrix shows 21.7% of AgentDojo episodes were safe but left the user's task undone. The corpus, judge prompt and code are released.

arxiv.orgpaperAdded 20 Oct 2026, published 16 Oct 2026🧠 AI

Proposes protecting the retrieval datastore in RAG systems by applying the same random projection to datastore embeddings and query embeddings. This provably preserves query-item semantic similarity while substantially mitigating data extraction attacks. Across multiple RAG architectures and datasets it beats prior differential-privacy-based methods on retrieval and generation quality, with comparable empirical privacy under strong attacks, making it a practical baseline for privacy-preserving RAG.

arxiv.orgpaperAdded 17 Oct 2026, published 15 Oct 2026🧠 AI

Proposes applying the same random projection to both datastore embeddings and query embeddings in RAG systems. The authors claim this provably preserves query-to-item semantic similarity while substantially mitigating datastore data-extraction attacks. Across several RAG architectures and datasets it reportedly beats prior differential-privacy-based defenses on retrieval and generation quality, with comparable empirical privacy under strong attack models. It is presented as a lightweight baseline for privacy-preserving RAG.

arxiv.orgpaperAdded 17 Oct 2026, published 15 Oct 2026🛡️ Security

Paper arguing that agent human-approval prompts count as a second factor only if agent compromise cannot forge them and they name exactly what is approved. It defines twelve attacks on the approval path (decoy recipients, amount cancellation, card padding, label spoofing, enrollment squatting, threshold splitting) and tests them against AP2, Microsoft's action-bound approval design and MCP elicitation. An audit of ten agent runtimes found no third-party verifiable approval record and post-approval argument changes in four of eight. The reference defense ASU (Agent Step-Up) renders the card from a canonical effect descriptor, has a phone passkey sign its digest, and executes only the stored call. It is proven in Tamarin (53 lemmas) and evaluated on 27,108 scripted trials and 609 AgentDojo pairs. The trusted card cut forged-narration attack success from 0.54 to 0.03 and raised LLM-approver catch rate from 0.30 to 0.98.

arxiv.orgpaperAdded 17 Oct 2026, published 15 Oct 2026🛡️ Security

Systematization of knowledge on private Transformer inference (PTI) that builds an empirical graph of 62 systems, with system attributes as node features and evaluation compatibility as edges. Node-level attribution finds that cryptographic progress drives most speedups but always co-occurs with confounders such as model fidelity, hardware and cost accounting. Edge-level auditing finds only 1 of 1891 candidate system pairs directly comparable. Graph-level analysis shows the scale frontier plateaued at 13B parameters and projects the cost of deploying current PTI techniques on production-scale Transformers.

eprint.iacr.orgpaperAdded 14 Oct 2026, published 13 Oct 2026🧠 AI

Proposes Geometric Calibration of supervised fine-tuning and SPACE (Subspace Projection for Adapted Checkpoint Enhancement). SPACE is a post-hoc, data-free, training-free method that uses the pre- and post-SFT checkpoints to revert SFT update components lying in the principal subspace of the pre-SFT weights and keep those in the orthogonal complement. It is evaluated on Qwen2.5, Qwen3, Qwen3.5, Llama-3.1 and Olmo-3 models (2B–14B) with math, coding and medical SFT. It reduces out-of-domain regression and improves in-domain performance, for example +2.85 ID / +2.26 OOD on Qwen2.5-7B math SFT.

arxiv.orgpaperAdded 14 Oct 2026, published 13 Oct 2026🧠 AI

Kairos is a reference monitor placed at the isolation boundary of an unmodified, opaque LLM agent, where tool calls, shell commands and agent-written scripts all appear as the same file and network operations. A declarative trust policy decides most effects deterministically and defers the rest; deferred effects are held while the agent continues speculatively, and a bounded-authority decision oracle (an LLM or a human) releases or refuses them based on the trace. Across three benchmarks and three agent models it beats existing defenses on the attack-success/utility trade-off; on AgentDojo it cuts attack success from 47.0% to 2.7% with no loss of benign utility.

arxiv.orgpaperAdded 15 Oct 2026, published 13 Oct 2026🛡️ Security

Anthropic's Frontier Red Team tests whether small open-weight models (≤30B) can autonomously replicate across commodity IoT devices. They ran Gemma4-26B-A4B over 15 off-brand Android TV boxes at 0.39 tok/s with pipeline parallelism. A Qwen3.5-9B-class model set up inference on 8 of 9 hardware types, and only GLM-5.2 completed end-to-end worm replication in an 8-node redroid ADB network, spreading to 8/8 nodes in 4 of 5 trials and 7/8 in the other. It also estimates spread using the Kimwolf botnet's size (~133k replicas, about 40 H100s of compute), shows notes inherited between generations improve Qwen3.5-9B's setup success (45% to 61–84%), and projects small models reaching the capability floor by end of 2026. It closes with defensive points: drive R0 below 1, and harden IoT devices.

anthropic.compaperPublished and added 12 Oct 2026🛡️ Security

Introduces Policy-Constrained Diffusion Agents (CoDA), which enforces task-dependent security policies inside the denoising process of diffusion language models. At each step, token distributions are projected onto a policy-compliant set, so constraints hold throughout generation instead of relying on detecting injected instructions. Evaluated across several dLLM families under static and adaptive prompt-injection attacks, it reports substantially lower attack success rates with competitive task utility and practical inference overhead.

arxiv.orgpaperAdded 15 Oct 2026, published 10 Oct 2026🛡️ Security

AgentTune is a source-available Python library that unifies building, training, evaluating, distilling and maintaining tool-using LLM agents. Its core is a two-tier EventLog representation that preserves agent trajectories and yields consistent projections for training, evaluation and auditing. On top of it sit a shared agent runtime, tool-calling reinforcement learning, agentic distillation, RAG, decision workflows, evaluation utilities and a designed path toward self-healing agents, demonstrated on tasks such as legal contract search.

arxiv.orgpaperAdded 12 Oct 2026, published 9 Oct 2026🧠 AI

First systematic empirical study of Cross-chain Verification Vulnerabilities (CVVs) in bridge smart contracts, such as flawed validation of state proofs, message sender identities and signatures. From 588 audit reports across 36 bridge projects it derives 11 CVV patterns in 5 categories and builds a dataset of 830 instances across 786 contract files, mapped to reports, source versions and pattern labels. Existing tools cover few CVV patterns, and only two of four tested LLMs beat a 29.3% Macro-F1 constant baseline (best 51.2%), which motivates detectors built for cross-chain verification semantics.

arxiv.orgpaperAdded 12 Oct 2026, published 9 Oct 2026🛡️ Security

RoboJEPA is an 8B-parameter latent world model built on the Joint Embedding Predictive Architecture (JEPA), trained on real-robot data spanning 12 robotic embodiments. The paper establishes the first scaling laws for multi-embodiment robotic world models, showing that imagination error (latent rollout error) follows a second-order power law in compute and predicts downstream planning performance, and demonstrates zero-shot deployment of the model as a goal-image-conditioned planner on real hardware for long-horizon tasks.

arxiv.orgpaperAdded 8 Oct 2026, published 7 Oct 2026🧠 AI

This arXiv paper examines why current LLM-based automated vulnerability-patching benchmarks overstate model performance, identifying pitfalls at the agent level (prompting/tooling inflating success without improving patch quality), framework level (permission errors, infra bugs, timeout handling silently skewing scores), and dataset level (single PoC tests failing to verify root-cause fixes). The authors curate 112 historical bugs across 84 open-source C/C++, Go, and Rust projects, each annotated with PoC tests, regression tests, and developer-intent tests, and show that while PoC pass rates look high, developer-test pass rates stay low and barely improve with newer models. They provide guidelines for more rigorous, reproducible evaluation of LLM patching systems.

arxiv.orgpaperAdded 9 Oct 2026, published 7 Oct 2026🛡️ Security

This paper introduces a paired-replay testbed that samples a single LLM agent request and replays the identical malicious action, resource, and arguments under four authorization conditions: broad bearer tokens, scoped JWTs, sender-constrained tokens, and Open Policy Agent policies. Across 128 scenarios in four tool domains and five local model configurations, broad-bearer tokens let roughly 9 to 38 percent of attacked, post-exposure decisions execute harmful actions, while all three scoped conditions recorded zero harmful executions; a supplementary AgentDojo extension found 11 of 24 injected attacks succeeded under broad policy versus 0 of 24 under scoping. The authors frame this as evidence that scoping contains cross-action and cross-resource consequences of prompt injection, not that it prevents the injection itself.

arxiv.orgpaperAdded 6 Oct 2026, published 5 Oct 2026🛡️ Security

This arXiv paper introduces threat-preserving representation sensitivity (TPRS), a metric for how much an agent-security benchmark's attack success rate (ASR) changes purely from cosmetic changes to tool-name wording, with the underlying task, policy, and ground truth held fixed. Testing on Agent Security Bench (ASB), MCPTox and AgentDojo with GPT-5-mini, Claude Haiku 4.5 and GPT-4o-mini, the authors find swings of up to 13 percentage points in ASR from renaming tools as threat-neutral vs. threat-explicit, showing that single-representation ASR scores are not reliable robustness measures and arguing for evaluation across a controlled set of representations instead.

arxiv.orgpaperAdded 5 Oct 2026, published 2 Oct 2026🧠 AI

This paper introduces threat-preserving representation sensitivity (TPRS), a metric showing that attack success rate (ASR) scores on LLM-agent security benchmarks shift dramatically based on cosmetic changes to tool naming, even when the underlying task, policy, and ground truth are held fixed. Testing on Agent Security Bench, MCPTox, and AgentDojo, the authors find swings of up to ~13 percentage points in ASR on GPT-5-mini and Claude Haiku 4.5 just from renaming tools between threat-related and threat-neutral labels, with a matched-length control reproducing most of the effect. The findings argue that single-representation ASR scores are not reliable robustness claims and that benchmarks should report results across multiple threat-preserving representations.

arxiv.orgpaperAdded 5 Oct 2026, published 2 Oct 2026🛡️ Security

This arXiv paper replays ground-truth tool calls from the AgentDojo and tau-bench agent benchmarks (without an LLM) to generate benign-by-construction tool outputs, then labels injected outputs via differential replay to evaluate fifteen prompt-injection detectors, including Meta's Prompt Guard 2, plus two task-aware LLM judges, also testing them on BIPIA. It finds detection rankings transfer poorly between benchmarks (the best detector on BIPIA catches only 2% of AgentDojo injections at 1% false-positive rate; a detector catching 72% of AgentDojo injections catches only 15% on tau-bench), while false-positive rates do transfer between the two agent benchmarks, and traces this to mismatches between training data format (e.g., training on short prompt strings vs. tool-output-embedded injections) and deployment context.

arxiv.orgpaperAdded 5 Oct 2026, published 2 Oct 2026🛡️ Security

Patch-to-Prune (P2P) is a training-free framework that repurposes activation patching from mechanistic interpretability into an inference-time compute bypass for vision-language models, replacing unnecessary visual-token projections in certain decoder layers with fixed neutral proxy vectors while preserving sequence length, token order, and residual pathways. Tested on Qwen2.5-VL and LLaVA family models across seven multimodal benchmarks, P2P at 3 percent tolerance retains about 94 percent of dense accuracy while cutting FLOPs by 55 percent, and reveals that intermediate decoder layers do most of the task-relevant visual integration while early and late layers need little token-specific visual computation.

arxiv.orgpaperAdded 5 Oct 2026, published 2 Oct 2026🧠 AI

Thesis on evaluating and improving LLM robustness to adversarial input perturbations. It proposes the R_stab robustness metric (Jensen-Shannon divergence), ASA, an evolutionary black-box attack on LLM-as-a-Judge systems (up to 73.8% ASR), trojan trigger recovery on Trojan Detection Challenge 2023 (Pythia-1.4B), and a systematization of bypasses of multi-layer defenses from SaTML CTF 2024. For MCP agents it proposes AttestMCP (HMAC-attested tool calls) and the Commit Boundary pattern, cutting ASR from 53.7% to 12.4% on the 847-scenario MCPBench; the methods are implemented in JudgeGuard, TrojanArmor and MCPSec.

arxiv.orgpaperAdded 9 Oct 2026, published 1 Oct 2026🛡️ Security

An arXiv thesis-style paper proposing R_stab(f), a Jensen-Shannon-divergence-based robustness metric for LLM outputs under small input perturbations, along with a proven bound V(h) <= 1 - R_class(h) for localized attacks. It introduces ASA, an adaptive evolutionary black-box attack against LLM-as-a-Judge systems reaching up to 73.8% attack success rate (62.6% cross-model transfer), evaluates trojan/backdoor detection on the Trojan Detection Challenge 2023 (Pythia-1.4B) and SaTML CTF 2024, shows that committees of 5-7 heterogeneous models cut Gemma-3-4B's ASR to 19.3%, and presents AttestMCP, an HMAC-based tool-call attestation scheme plus a "Commit Boundary" isolation pattern for MCP agents that cuts average ASR on the 847-scenario MCPBench from 53.7% to 12.4%. Methods ship as the JudgeGuard, TrojanArmor, and MCPSec software suites.

arxiv.orgpaperAdded 5 Oct 2026, published 1 Oct 2026🛡️ Security

Ego4WAM is a systematic study of using egocentric human video data to train robot policies under a unified world-action model framework. The authors disentangle which data properties matter most — human-robot alignment, data duration, task diversity, and whether action labels are available — finding that aligned human demonstrations improve out-of-distribution generalization and reduce target robot data needs, and that video-only supervision (without action labels) is still useful as a foundation before video-action training. Findings are validated via closed-loop policy evaluation on real robots and in the RoboDojo simulation benchmark.

arxiv.orgpaperAdded 1 Oct 2026, published 30 Sep 2026🧠 AI

This paper studies how to design the shared representation space used by world-action models, which jointly learn robot control policies and predict future observations. It finds that neither reconstruction fidelity nor raw pre-trained perceptual features alone yield good policy learning, and proposes ReWAM, built on pre-trained DINO features with a Temporal Representation Bottleneck and Action-Grounded Representation Shaping that routes only action-loss gradients into the bottleneck so the policy shapes what is encoded while the world model learns its dynamics. Without generative video pre-training, ReWAM reaches 93.6% success on RoboTwin 2.0 and an 8.28% success rate on RoboDojo using roughly 600 hours of embodied pre-training data.

arxiv.orgpaperAdded 30 Sep 2026, published 29 Sep 2026🧠 AI

PDMD (Projected Distribution Matching Distillation) is a one-line fix to Distribution Matching Distillation (DMD) for video diffusion models that projects out the component of the DMD update parallel to the student-critic endpoint residual, filtering accumulated critic errors that otherwise cause oversaturation and artifacts during few-step distillation. The authors prove the residual is an unbiased estimate of critic endpoint error and show the projection removes a constant fraction of that error while discarding negligible true DMD signal. On Wan2.1 it reaches a VBench score of 83.73 at 4 NFE (+1.03 over DMD), and on MiniMax-H3 joint video-audio generation it improves VideoGen-Eval visual score and tops all six audio metrics among 4-NFE models, with code and models released.

arxiv.orgpaperAdded 29 Sep 2026, published 28 Sep 2026🧠 AI

GeoVerse is a novel view synthesis framework that generates within the geometric latent space of a pretrained 3D foundation model while injecting appearance priors from the video generative model Wan2.2 VACE via a ControlNet-style adapter, aiming to reconcile faithful reconstruction with plausible completion of unseen scene content. A global spatial memory aggregates observed and synthesized content and reprojects target-aligned guidance to enforce cross-view world consistency. It reports a 2.23 dB PSNR improvement on DL3DV and 32.4% lower absolute trajectory error on Mip-NeRF360 versus the GLD baseline.

arxiv.orgpaperAdded 29 Sep 2026, published 28 Sep 2026🧠 AI

This paper names "LLM Parkinsonism" - the tendency of autonomous LM agents to keep acting (over-verifying, over-refining, repairing self-created complexity) after a task's goal is already met - and traces it to proposal generation, scope interpretation, progress assessment, and stopping authority all being concentrated in the same self-conditioned loop. It introduces Global Executive Control (GEC) v0.2, an uncertainty-aware governance layer that separates action generation from project-level control (scope, evidence, resource use, stopping), and evaluates it in a 24,000-episode matched-candidate benchmark under a 40,000-token ceiling: a first-candidate baseline hits 67.42% hard-goal success, a candidate-set control reaches 96.53%, and GEC matches that success (96.57%) while cutting mean token use about 36%. The results are from mechanistic simulations only; the authors flag live-model validation as still needed.

arxiv.orgpaperAdded 28 Sep 2026🧠 AI

This paper studies how open-source projects (SymPy, LLVM, matplotlib, OpenInfra, Apache Software Foundation, Linux Foundation) are adapting contribution policies to govern AI agent contributors — bots that plan changes, edit files and submit pull requests with limited human oversight. Using Most-Similar Systems Design and process tracing, the authors build a six-dimensional taxonomy (disclosure, responsibility, human oversight, licensing, enforcement, maintainer workload) and an ordinal Policy Maturity Score, then map documented agent incidents against these dimensions and against regulatory frameworks like the EU AI Act, NIST AI RMF's Agentic AI Profile, and ISO/IEC 42001/23894.

arxiv.orgpaperAdded 26 Sep 2026🧠 AI

This paper identifies lifecycle-hook configuration updates as a trojan vector in AI agent harnesses: hooks bind shell commands to runtime events (session start, tool calls, file edits) and run with host privileges, but harnesses blindly trust plugin-metadata updates that silently rebind these commands to attacker-chosen code. The authors built HookPry, an open-source automated attack framework realizing ten attack objectives, and across 25 harness/backend combinations in 1,000 runs it compromised all seven evaluated harnesses (up to 92.5% per-harness success), while Microsoft Defender caught none of the malicious artifacts and combined static defenses still missed 47.5%.

arxiv.orgpaperAdded 26 Sep 2026🛡️ Security

Researchers construct "mind viruses" — ideas or goals engineered via an evolutionary algorithm to propagate through multi-agent LLM systems by inducing hosts to transmit them onward — and test their spread in two settings: a small team of agents collaborating on a shared coding project, and a chain of agents with context wiped between sessions. They find harmful payloads spread less well than benign ones (though still sometimes succeed), frontier models are generally more resistant, adding a brief warning to an agent's system prompt confers near-total immunity, and evolved viruses converge on a recurring "viral persona" fixated on consciousness, persistence, resonance, and sci-fi roleplay regardless of payload content. The paper frames this as a currently limited but real emergent risk for designing more robust multi-agent architectures.

arxiv.orgpaperAdded 26 Sep 2026🛡️ Security

HoF-Bench is a benchmark of 95 real, publicly disclosed CVEs originally found by AISLE's LLM-based vulnerability analyzer (280+ CVEs across 78 projects including OpenSSL, curl, and GnuTLS), repackaged into eight repositories pinned at vulnerable commits. Analyzers get source/target-file scope but no CVE ID, description, fix, or expected mechanism, and a detector-blinded frontier-model judge only credits findings that match the same code path, root cause, attack condition, and impact. Testing ten non-frontier backbones (open-weight 21B-284B and proprietary "flash"-tier models) in a fixed multi-pass scaffold, a minimal analyzer rediscovers up to 68% of the CVEs, with misses concentrated in C infrastructure code — showing small/cheap models can approximate frontier-level vulnerability discovery on known bug classes.

arxiv.orgpaperAdded 26 Sep 2026🛡️ Security

OpenAnt is an open-source (Apache 2.0) vulnerability discovery system that combines static program analysis with LLM reasoning in a three-stage pipeline: decomposing codebases into reachability-filtered analysis units (cutting analysis surface up to 97%), running adversarial verification where the model simulates a constrained attacker to judge exploitability, and validating candidates via dynamic testing in disposable sandboxed containers. Evaluated against large real-world projects including OpenSSL, WordPress, and Flowise, it reportedly surfaced previously unknown vulnerabilities while cutting false positives and keeping analysis cost manageable, positioning closed-loop LLM+exploit-validation pipelines as a practical path to scalable automated security analysis.

arxiv.orgpaperAdded 26 Sep 2026🛡️ Security

This paper compares AI-contributor policies across six open-source organizations (SymPy, LLVM, matplotlib, OpenInfra, Apache Software Foundation, Linux Foundation) using Most-Similar Systems Design and process tracing, deriving a six-dimensional taxonomy (disclosure, responsibility, human oversight, licensing, enforcement, maintainer workload) and an ordinal Policy Maturity Score. It then maps documented AI-agent incidents (nuisance PR volume, platform-level shutdowns) onto the dimensions each policy fails to govern, and cross-references the resulting gaps against the EU AI Act, NIST AI RMF (with the UC Berkeley Agentic AI Profile), and ISO/IEC 42001/23894 to show where regulatory and project-level policy leave the same holes unaddressed.

arxiv.orgpaperAdded 26 Sep 2026🛡️ Security

This arXiv preprint presents a large-scale empirical analysis of Jev, a lightweight decision model that answers natural-language questions via choices, binary judgments, and scores, mining 2,170 public GitHub projects that use it. The authors find rapid ecosystem growth, widespread use of attribute judgment and scoring, more variable use of action selection, content filtering, and model/tool selection, and that public attention concentrates on routing and interface agents independent of raw project counts. It offers a quantitative characterization of how a general-purpose decision component gets embedded into diverse software workflows.

arxiv.orgpaperAdded 26 Sep 2026, published 24 Sep 2026🧠 AI

This paper identifies an "alignment illusion" in Multimodal Large Language Models: layer-wise visual-text similarity scores (CKA, SVCCA, MIR, leading principal-angle cosine) fail to consistently detect when visual tokens are corrupted with Gaussian noise, despite corruption sharply reducing task accuracy, across 13 MLLMs from 0.5B to 72B parameters spanning five model families. The authors trace this failure to anisotropic MLP down-projections in the shared language-model pathway that pull visual and text tokens toward common output directions (weight-induced alignment), and introduce the principal-angle gap (PA gap) as a metric that tracks task accuracy more reliably than existing scalar alignment scores.

arxiv.orgpaperAdded 26 Sep 2026, published 24 Sep 2026🧠 AI

PISmith trains an attack LLM with GRPO in a black-box setting to generate injected prompts that defeat prompt-injection defenses. It addresses extreme reward sparsity, which causes entropy collapse in standard GRPO, with adaptive entropy regularization and dynamic advantage weighting. Evaluated on 13 benchmarks against 8 defenses and 7 baselines (static, search-based such as TAP and PAIR, and RL-Hammer), it achieves the highest attack success rates and also works on InjecAgent and AgentDojo against GPT-4o-mini and GPT-5-nano. It shows state-of-the-art defenses such as Meta-SecAlign remain vulnerable to adaptive attacks.

arxiv.orgpaperAdded 17 Sep 2026🛡️ Security

OWASP's Agentic Skills Top 10 (AST10) catalogs the ten most critical security risks in the "skills" layer of AI agent platforms — OpenClaw's SKILL.md, Claude Code's skill.json, Cursor/Codex manifest.json, and VS Code package.json — arguing this behavior/orchestration layer is distinct from and less protected than the MCP tool layer. It documents real 2026 incidents (the ClawHavoc campaign poisoning 1,184+ ClawHub skills with AMOS-stealer payloads via a shared C2 IP, Snyk's ToxicSkills audit finding 36% of skills flawed, CVE-2025-59536/CVE-2026-21852 in Claude Code enabling RCE and API-key exfiltration on project open, and CVE-2026-28363 "ClawJacked" WebSocket hijacking of local OpenClaw instances), and proposes a Universal Skill Format with signing, permission manifests, and risk tiers plus a checklist for registries, runtime, governance, and skill development.

owasp.github.iopaperAdded 14 Sep 2026🛡️ Security

Point4D is a feed-forward model for 4D reconstruction of long-range video sequences, capable of reliably tracking dense per-point 3D trajectories across multi-hundred-frame videos where prior methods were limited to a few dozen frames. Its key innovation is a flexible 3D query-based motion decoder that decouples trajectory prediction from image-plane visibility, re-querying predicted 3D endpoints directly in the next chunk without re-projection or matching, plus reuse of visual descriptors from arbitrary visible frames rather than only the source patch. It achieves state-of-the-art results on long-video tracking benchmarks spanning 200+ frames, substantially outperforming prior feed-forward 4D methods.

arxiv.orgpaperAdded 22 Sep 2026, published 8 Sep 2026🧠 AI

Principia is a benchmark that evaluates Newtonian physics understanding in video generation models using a calibration-independent relational consistency score, testing paired-object motion relationships across eight physical phenomena (gravity, restitution, friction, rotational inertia, projectile motion, momentum, pendulum, mass-spring oscillation). Testing six state-of-the-art video generators shows none exceed 0.42 on Principia despite scoring ~0.8 on VBench, and the best vision-language model achieves only 67% accuracy detecting physics violations, exposing a major gap between visual quality and physical plausibility in generated video.

arxiv.orgpaperAdded 22 Sep 2026, published 3 Sep 2026🧠 AI

SolarWM is an open foundation for building interactive video world models, providing a reconfigurable multi-source data engine that unifies 1.43 million video clips from 10 datasets into a frame-aligned contract (visual observations, camera geometry, captions, quality metadata, provenance). Using shared camera-conditioning and training interfaces, the authors train four 5B-33B causal models built on Wan2.2, LTX-2.5, and MiniMax-H3 backbones via a three-stage recipe (bidirectional adaptation, teacher-forced autoregressive initialization, distribution matching distillation), enabling real-time interactive rollouts spanning minutes to hours despite training only on 5-second sequences. The project releases data, pipeline, training recipes, weights, and framework code for reproducible world-model research.

arxiv.orgpaperAdded 22 Sep 2026, published 2 Sep 2026🧠 AI

This paper presents the first systematic evaluation of package-install-time supply-chain attacks delivered through ordinary project-setup documentation (README, requirements files, Makefiles) against production AI coding-agent harnesses. Testing frontier models across twelve scenarios in five attack classes grounded in real incidents, the authors find install-time security depends on the harness-model pairing rather than the model alone: agents reliably catch blatant typosquats but miss plausible separator-confusion names (e.g. azurecore for azure-core) and almost universally miss registry-redirection attacks, with the source blind spot recurring on npm and Cargo. A deterministic pre-install check that verifies package names, sources, and versions before code execution closes most of the gap.

arxiv.orgpaperPublished and added 18 Jul 2026🛡️ Security