Like a Hammer, It Can Build, It Can Break: Large Language Model Uses, Perceptions, and Adoption in Cybersecurity Operations on Reddit

Large language models LLMs have recently emerged as promising tools for augmenting Security Operations Center SOC workflows, with vendors increasingly marketing autonomous AI solutions for SOCs. However, there remains a limited empirical understanding of how such tools are used, perceived, and...

OpenClaw 安全漏洞

OpenClaw is an open-source intelligent artificial assistant developed by OpenClaw. Versions of OpenClaw prior to 2026.3.24 contained security vulnerabilities. These vulnerabilities stemmed from authorization bypasses in the HTTP/v1/models endpoints, which failed to enforce the requirement for...

EUVD-2026-20928

A security flaw has been discovered in GL.iNet GL-RM1, GL-RM10, GL-RM10RC and GL-RM1PE 1.8.1. Affected by this issue is some unknown functionality of the component Factory Reset Handler. Performing a manipulation results in improper authentication. The attack can be initiated remotely. The...

ONNX: Malicious ONNX models can crash servers by exploiting unprotected object settings.

...

wasmtime 安全漏洞

Wasmtime is a lightweight WebAssembly runtime open source by the Bytecode Alliance. Versions prior to 24.0.7, 36.0.7, 42.0.2, and 43.0.1 of Wastime contain security vulnerabilities. These vulnerabilities arise when using the Val type promotion feature to increase values of component models. If bi...

CVE-2026-5815

A vulnerability was detected in D-Link DIR-645 1.01/1.02/1.03. Impacted is the function hedwigcgimain of the file /cgi-bin/hedwig.cgi. The manipulation results in stack-based buffer overflow. The attack can be launched remotely. The exploit is now public and may be used. This vulnerability only...

Allocation of Resources Without Limits or Throttling

Overview react-server-dom-parcel is a React Server Components bindings for DOM using Parcel. This is intended to be integrated into meta-frameworks. It is not intended to be imported directly. Affected versions of this package are vulnerable to Allocation of Resources Without Limits or Throttling...

Allocation of Resources Without Limits or Throttling

Overview react-server-dom-turbopack is a React Server Components bindings for DOM using Turbopack. This is intended to be integrated into meta-frameworks. It is not intended to be imported directly. Affected versions of this package are vulnerable to Allocation of Resources Without Limits or...

LobeHub: Unauthenticated authentication bypass on `webapi` routes via forgeable `X-lobe-chat-auth` header

Summary The webapi authentication layer trusts a client-controlled X-lobe-chat-auth header that is only XOR-obfuscated, not signed or otherwise authenticated. Because the XOR key is hardcoded in the repository, an attacker can forge arbitrary auth payloads and bypass authentication on protected...

GHSA-5MWJ-V5JW-5C97 LobeHub: Unauthenticated authentication bypass on `webapi` routes via forgeable `X-lobe-chat-auth` header

Summary The webapi authentication layer trusts a client-controlled X-lobe-chat-auth header that is only XOR-obfuscated, not signed or otherwise authenticated. Because the XOR key is hardcoded in the repository, an attacker can forge arbitrary auth payloads and bypass authentication on protected...

PT-2026-31562

Open AI models detect vulnerabilities by analyzing code, configs, and system behavior step-by-step—like spotting the FreeBSD zero-day CVE-2024-47467 in the Mythos showcase. Even a 3B open model nailed it, matching bigger ones. The chart shows rankings flip across tasks: OWASP easy stuff vs. gnarl...

arthexis (>=0.2.6 <=0.8.0), cg-django-uaa (=2.1.9) +29 more potentially affected by CVE-2026-3902 via django (>=5.2.0 <=5.2.12)

django PYPI version =5.2.0, =0.2.6, =0.1.0, =0.1.0, =1.3.0, =1.92.0.5, =4.2.0, =0.0.7, =3.0.0, =0.1.0, =0.1.1 and more Source cves: CVE-2026-3902 Source advisory: OSV:GHSA-MVFQ-GGXM-9MC5...

EUVD-2026-19671

text-generation-webui is an open-source web interface for running Large Language Models. Prior to 4.3, he superbooga and superboogav2 RAG extensions fetch user-supplied URLs via requests.get with zero validation — no scheme check, no IP filtering, no hostname allowlist. An attacker can access clo...

From Incomplete Architecture to Quantified Risk: Multimodal LLM-Driven Security Assessment for Cyber-Physical Systems

Cyber-physical systems often contend with incomplete architectural documentation or outdated information resulting from legacy technologies, knowledge management gaps, and the complexity of integrating diverse subsystems over extended operational lifecycles. This architectural incompleteness...

Guiding Symbolic Execution with Static Analysis and LLMs for Vulnerability Discovery

Symbolic execution detects vulnerabilities with precision, but applying it to large codebases requires harnesses that set up symbolic state, model dependencies, and specify assertions. Writing these harnesses has traditionally been a manual process requiring expert knowledge, which significantly...

SkillSieve: A Hierarchical Triage Framework for Detecting Malicious AI Agent Skills

OpenClaw's ClawHub marketplace hosts over 13,000 community-contributed agent skills, and between 13% and 26% of them contain security vulnerabilities according to recent audits. Regex scanners miss obfuscated payloads; formal static analyzers cannot read the natural language instructions in...

CritBench: A Framework for Evaluating Cybersecurity Capabilities of Large Language Models in IEC 61850 Digital Substation Environments

The advancement of Large Language Models LLMs has raised concerns regarding their dual-use potential in cybersecurity. Existing evaluation frameworks overwhelmingly focus on Information Technology IT environments, failing to capture the constraints, and specialized protocols of Operational...

Stop Fixating on Prompts: Reasoning Hijacking and Constraint Tightening for Red-Teaming LLM Agents

With the widespread application of LLM-based agents across various domains, their complexity has introduced new security threats. Existing red-team methods mostly rely on modifying user prompts, which lack adaptability to new data and may impact the agent's performance. To address the challenge,...

Swiss-Bench 003: Evaluating LLM Reliability and Adversarial Security for Swiss Regulatory Contexts

The deployment of large language models LLMs in Swiss financial and regulatory contexts demands empirical evidence of both production reliability and adversarial security, dimensions not jointly operationalized in existing Swiss-focused evaluation frameworks. This paper introduces Swiss-Bench 003...

Hackers or Hallucinators? A Comprehensive Analysis of LLM-Based Automated Penetration Testing

The rapid advancement of Large Language Models LLMs has created new opportunities for Automated Penetration Testing AutoPT, spawning numerous frameworks aimed at achieving end-to-end autonomous attacks. However, despite the proliferation of related studies, existing research generally lacks...