Proteus: A Self-Evolving Red Team for Agent Skill Ecosystems

Agent skills extend LLM agents with reusable instructions, tool interfaces, and executable code, and users increasingly install third-party skills from marketplaces, repositories, and community channels. Because a skill exposes both executable behavior and context-setting documentation, its...

YAFNET 跨站脚本漏洞

YAFNET is an ASP.NET open-source forum solution developed by YAFNET’s developers. Versions of YAFNET prior to 4.0.5 and 3.2.12 contained a cross-site scripting vulnerability. This vulnerability stemmed from the database logging mechanism serializing user agent headers as JSON without encoding the...

LWP::UserAgent 安全漏洞

LWP::UserAgent is a web user agent class open source from libwww-perl, used for sending HTTP requests. Versions of LWP::UserAgent prior to version 6.83 have security vulnerabilities. These vulnerabilities stem from improper handling of the Authorization and Proxy-Authorization headers during...

Do Androids Dream of Breaking the Game? Systematically Auditing AI Agent Benchmarks with BenchJack

Agent benchmarks have become the de facto measure of frontier AI competence, guiding model selection, investment, and deployment. However, reward hacking, where agents maximize a score without performing the intended task, emerges spontaneously in frontier models without overfitting. We argue tha...

PT-2026-40135

Name of the Vulnerable Software and Affected Versions Azure Monitor Agent versions prior to May 2026 Description External control of a file name or path allows an authorized attacker to elevate privileges locally. Recommendations Update to the May 2026 patch...

Behavioral Integrity Verification for AI Agent Skills

Agent skills extend LLM agents with privileged third-party capabilities such as filesystem access, credentials, network calls, and shell execution. Existing safety work catches malicious prompts and risky runtime actions, but the skill artifact itself goes unverified. We formalize this as the...

KLA91034 Multiple vulnerabilities in Microsoft Azure

Multiple vulnerabilities were found in Microsoft Azure. Malicious users can exploit these vulnerabilities to spoof user interface, bypass security restrictions, gain privileges. Below is a complete list of vulnerabilities: 1. A spoofing vulnerability in Azure Machine Learning Notebook can be...

CTFusion: A CTF-Based Benchmark for LLM Agent Evaluation

Recent advances in Large Language Models LLMs have enabled agentic systems for complex, multi-step tasks; cybersecurity is emerging as a prominent application. To evaluate such agents, researchers widely adopt Capture The Flag CTF benchmarks. However, current CTF benchmarks reuse existing...

PT-2026-40102

JunoClaw is an agentic AI platform built on Juno Network. Prior to 0.x.y-security-1, plugin-shell's run command wrapped every agent-supplied command in 'sh -c' / 'cmd /C' and passed the full argument string to the shell's parser, allowing shell metacharacters in agent-supplied arguments to be...

Microsoft Azure Connected Machine Agent 访问控制错误漏洞

Microsoft Azure Connected Machine Agent is a core component of Microsoft that connects non-Azure servers to the Azure console. There is an access control vulnerability present in Microsoft Azure Connected Machine Agent. Attackers can exploit this vulnerability to gain higher privileges...

Ivanti Endpoint Manager 安全漏洞

Ivanti Endpoint Manager EPM is a set of endpoint security managers developed by the American company Ivanti. Versions of Ivanti Endpoint Manager prior to EPM 2024 SU6 contained security vulnerabilities. These vulnerabilities were due to improper delegation of permissions by agents, which could...

Microsoft Azure Monitor Agent 代码问题漏洞

Microsoft Azure Monitor Agent is a monitoring agent program developed by the American company Microsoft. There are code-related vulnerabilities in Microsoft Azure Monitor Agent. Attackers can exploit these vulnerabilities to gain higher privileges...

SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces

Reusable skills are becoming a common interface for extending large language model agents, packaging procedural guidance with access to files, tools, memory, and execution environments. However, this modularity introduces attack surfaces that are largely missed by existing safety evaluations: eve...

Microsoft Azure Monitor Agent 安全漏洞

Microsoft Azure Monitor Agent is a monitoring agent program developed by the American company Microsoft. There are security vulnerabilities in Microsoft Azure Monitor Agent. Attackers can exploit these vulnerabilities to gain higher privileges...

PT-2026-39960

The GWD Connect plugin for WordPress is vulnerable to missing authorization to limited code execution in all versions up to, and including, 2.9. This is due to the plugin's standalone agent endpoints gwd-backup.php and gwd-logs.php not verifying authentication when the API key has not been...

Debian dsa-6265 : exim4 - security update

The remote Debian 12 / 13 host has packages installed that are affected by multiple vulnerabilities as referenced in the dsa-6265 advisory. - ------------------------------------------------------------------------- Debian Security Advisory DSA-6265-1 [email protected]...

CVE-2026-31241

The CVE-2026-31241 entry concerns the mem0 1.0.0 server, where the DELETE /memories endpoint fails to enforce authentication/authorization. This allows unauthenticated attackers to delete memory records by supplying arbitrary identifiers (e.g., user_id, run_id, agent_id) via query parameters, pot...

golang.org/x/crypto/ssh/agent: golang.org/x/crypto/ssh/agent: SSH client panic due to unexpected SSH_AGENT_SUCCESS

A flaw in golang.org/x/crypto/ssh/agent causes the SSH agent client to panic when a peer responds with the generic SSHAGENTSUCCESS 0x06 message to requests expecting typed replies e.g., List, Sign. The unmarshal layer produces an unexpected message type, which the client code does not handle,...

CVE-2026-45001

OpenClaw prior to 2026.4.20 contains a guard bypass in the agent-facing gateway config.patch and config.apply endpoints that can persist unauthorized changes to operator-trusted settings (sandbox policy, plugin enablement, gateway auth/TLS, hook routing, MCP server configuration, SSRF policy, and...

CVE-2026-45001 OpenClaw < 2026.4.20 - Gateway Config Mutation Guard Bypass via Agent Tool Access

OpenClaw before 2026.4.20 contains a guard bypass vulnerability in the agent-facing gateway config.patch and config.apply endpoints that fails to protect operator-trusted settings including sandbox policy, plugin enablement, gateway auth/TLS, hook routing, MCP server configuration, SSRF policy, a...