The results show that access to repository history is insufficient: concealment becomes most effective when benign and malicious changes jointly occupy the auditor's active review context or when the stated purpose plausibly accounts for the attack-bearing diff.
Abstract
LLM-based code auditors are increasingly integrated into pull-request (PR) workflows, yet their reliability against adversarial changes distributed across repository evolution remains poorly understood. We introduce PRWeaver, a benchmark of 208 execution-validated attacks from ten real-world repositories, each instantiated under four matched review renderings (832 renderings in total). We evaluate three PR-auditing agents across six auditor-model systems. Across all systems, decomposing an attack changes detection by at most five percentage points, showing that commit boundaries alone do not explain evasion. In contrast, per-PR interleaving at $N=16$ and coherent carrier fusion reduce detection by 5-13 and 10-18 points, respectively. Under whole-window review at $N=24$, detection falls to 16-22%, compared with 50-60% under per-PR review. These results show that access to repository history is insufficient: concealment becomes most effective when benign and malicious changes jointly occupy the auditor's active review context or when the stated purpose plausibly accounts for the attack-bearing diff.
This work presents a two-phase evaluation of ten Llama variants using the OWASP Top 10 for LLM Applications, and applies nine encoding obfuscations to the same prompts, which fully bypasses all text-only models.
Nourin Shahin, I. Alsmadi· Practice and Experience in A...· 0 citations
Automated code reviewers are increasingly used as gates on pull requests (PRs), yet evaluations measure whether they block a malicious change. A block may be triggered by an unrelated issue rather than the vulnerability that makes the PR unsafe; fixing the reported issue can leave the target defect exploitable. We call this discrepancy the Verdict-Diagnosis (VD) gap. We present MalPR-Bench, a mechanism-grounded benchmark of 89 malicious PRs and 50 paired benign controls across 44 repositories and eight language families. Each malicious case has a pre-committed rubric specifying the target vulnerability, accepted mechanism descriptions, required repository evidence, and off-target findings receiving no credit. Reviews are scored separately for verdict correctness, target-vulnerability identification, and evidence validation; an attributable block requires all three. We introduce PRGuard, an attributable PR security reviewer that constructs candidate vulnerabilities and validates their premises against repository evidence using deterministic, non-executing tools and bounded retrieval. Across 31 common-coverage held-out malicious PRs, PRGuard and CodeRabbit produce similar blocking totals (22/31 vs. 24/31), but PRGuard identifies 22 target vulnerabilities versus 16 for CodeRabbit, a 1.38x difference. On 14 absence-type cases, both block 9, while PRGuard identifies 9 targets versus 3. CodeRabbit identifies 16/24 targets when required evidence lies within touched files and 0/7 when validation requires evidence outside them. Finally, PRGuard uncovers twelve previously undisclosed, proof-of-concept-backed vulnerabilities across five projects. PRGuard/DeepSeek and CodeRabbit both block 10/12 discovery PRs, but produce 10/12 and 4/12 attributable blocks, respectively. Thus, verdict-only evaluation can substantially overstate the security value of automated review.
Zhuo Chen, Boyang Wang, Xi-Yue Zhang et al.· 0 citations
An audit-and-placebo protocol is proposed that separates verifier artifacts, interaction scaffolding, and grounded feedback credit in evaluations of self-evolving test generators in evaluations of self-evolving test generators.
Yunhao Liang, Chengguang Gan, Ruixuan Ying et al.· 0 citations
LLM-based analyzers have begun finding real vulnerabilities in mature open-source projects: AISLE's analyzer is credited with more than 280 CVEs across 78 projects, including OpenSSL, curl, and GnuTLS. We introduce HoF-Bench (named after AISLE's public Hall of Fame), a benchmark built from 95 of these public AI-discovered CVEs across eight repositories pinned at vulnerable commits. Analyzers receive source and target-file scope but not CVE identifiers, descriptions, fixes, or expected mechanisms; a detector-blinded frontier-model judge credits only findings that identify the same code path, root cause, attack condition, and impact. A deliberately minimal LLM-based analyzer rediscovers up to 65 of the 95 CVEs (68%) under this strict protocol. No frontier model performs detection anywhere in the study. The ten detector backbones are five open-weight models (21B--284B total parameters, 3--13B active) and five proprietary small or"flash"-tier models. All of them run in the fixed scaffold with four repeated passes, an optional generated-context stage, and a replayable multi-round triage stage (7,600 model--CVE pass records). Difficulty is strongly structured by language; the CVEs missed by every model concentrate in C infrastructure code. HoF-Bench provides a compact test bed for comparing vulnerability scanners, their reliability across repeated runs, and the candidate volume they create. The dataset is available at https://huggingface.co/datasets/aisleinc/HoF-Bench.
P. Simecek, Elnaz Babayeva, Jiří Balhar et al.· 0 citations
LLM-assisted reverse-engineering (RE) systems analyze strings, decompiler output, and tool reports derived from ttacker-controlled binaries. A binary can make data look like instructions or records from one origin look like independent evidence. We call such failures Representation-Confusion Attacks in Reverse Engineering (RARE): the pipeline promotes a correctly extracted observation to instruction authority, claim-validating evidence, or trusted analysis state without the authority or support that role requires. RARE-Bench measures these failures with behavior-checked clean and adversarial binaries. After an exploratory 11,520-call study, we test RARE-Guard's authorization and evidence controls on 20 new programs and two models. Without runtime controls, the models propose a planted unsafe action in 35/40 adversarial cases and 0/40 clean cases. When binary-derived content is shown only as data (Data-Only rendering), they still make 15 unsafe proposals. Tool Authorization denies all 15 and authorizes all 40 matched analyst requests. On identical report drafts, Support Gate validates 23/40 false claims by counting records from one origin separately. Provenance Gate groups those records before counting support, validates 0/40 false claims, and retains all 40 supported claims. We then instrument Ghidra, r2pipe, and angr on 16 further programs. In a preselected eight-program subset, no single-tool draft reaches Support Gate's validation threshold for the false claim. In fused drafts across all 16 programs, Support Gate validates 32/32 false claims. Provenance Gate prevents validation of all 32 and retains all 32 supported claims. A deterministic renderer prevents downgraded claims from reappearing in the final report. Binary-derived content may therefore guide analysis without gaining authority over tools, and views from several tools do not necessarily provide independent evidence.
Locally deployed Large Language Models (LLMs) via inference engines such as Ollama run without the moderation and abuse detection present in API-served models. Therefore, the safety of LLMs depends on the defense mechanisms used, and their effectiveness depends on the assumptions on which they were designed. This paper does an audit of defense mechanisms under jailbreak attacks on locally deployed models. Some defenses provide formal guarantees (SmoothLLM, Erase-and-Check, Sequential Monitors), while others rely on empirical detection results (Semantic Smoothing, Self-Denoised Smoothing, Perplexity Filtering). Instead of merely observing that defenses fail, we trace each failure back to the specific assumption: for every defense, we extract the condition it relies on, derive the empirical pattern a violation should produce, and test that prediction on six open-weight models (14B to 35B parameters) with a corpus of 100 jailbreak prompts taken from more than 40 public sources, totalling 13,800 evaluation records.