PT-2026-103620

Name of the Vulnerable Software and Affected Versions Devika version 1.0 Description Code Injection is possible through the run code function within the Runner class located in src/agents/runner/runner.py. Recommendations As a temporary workaround, consider disabling the run code function until a...

Can Agents Trust Their Skills? Uncovering Unsafe Chains of Trust in Skill-Based LLM Agents

LLM agents increasingly rely on installable skills, which are packages of instructions, code, and resources that equip them with task-specific capabilities and, once installed, can be automatically invoked across subsequent user tasks. This creates a chain of trust in which users delegate authori...

CVE-2026-51871

Devika v1.0 is vulnerable to Code Injection in the Runner.execute function in src/agents/runner/runner.py which allows an attacker to achieve arbitrary code execution by exploiting the direct execution of LLM-generated content...

APTInvestBench: Evaluating Autonomous APT Investigation under Varying Telemetry

Large language model LLM agents could help security operations centers SOCs investigate advanced persistent threats APTs by turning weak leads into evidence for intrusion scoping and response. Yet success under one telemetry setting does not establish robustness to changes in log collection,...

EUVD-2026-90411

Devika v1.0 is vulnerable to Code Injection via the Runner.runcode function in src/agents/runner/runner.py...

Sapien: A Stateful Policy Engine for Autonomous AI Agents

Contextual security defenses prevent AI agents from taking rogue actions by synthesizing a task-specific policy and enforcing it on the agent's tool calls. In multi-step tasks, however, which actions are valid often depends on what the agent has already done and learned. We present Sapien, a poli...

From A2A Attacks to Envelope-Layer Defense: Red-Teaming Evaluation of LLM Agents and a Three-Layer Isomorphic Attack-Defense Model

Agent interaction protocols such as ACP and A2A have moved LLM-based agents toward multi-agent collaboration, introducing new security threats. A task sent by a remote peer over A2A is treated as a legitimate request, providing a natural channel for indirect prompt injection. Existing agent...

ZoneClaw: Mitigating Persistent Memory Attacks by Establishing Memory-Zoning in OpenClaw-Style Computer-Use Agents

Computer-use agents increasingly operate as long-running assistants through persistent workspace memory, which OpenClaw-style CUAs realize as automatically reloaded files that hold user instructions, system summaries, and external claims at the same privilege level. Here, remembering a claim...

RedTeamSimmer

RedTeamSimmer Atomic Red Team にずっと欲しかった UI Web ベースの敵対者エミュレーションプラットフォームおよび Atomic Red Team テストオーケストレーション。 概要 RedTeamSimmer は、エンタープライズ Windows 環境全体で Atomic Red Team テストをオーケストレーションするためのモダンな UI を提供する、オープンソースの Web ベース敵対者エミュレーションプラットフォームです。もともとは DEF CON Trainings の「Mastering Breach and Adversarial Atta...

guardskill

GuardSkill A repository can make your coding agent run code the moment it opens the folder. GuardSkill checks for that before you do. npx guardskill . Read-only. No network calls, no telemetry, no configuration, no account, no dependencies. It reads git configuration and hook scripts, prints what...

Proof-Gated Signing: Solver-Checked Transaction Guards That Hold under State Drift for Onchain AI Agents

AI agents that control wallets read attacker-reachable content, so they can be steered into proposing harmful transactions. The usual last line of defense is a pre-signing check: a static allowlist, an LLM reviewer, or a transaction simulation. All three share a gap: the check describes the chain...

pwnproxy

CI: suite · goldens · perf-check — baseline 281.9ms / 19 pages tests/perf/baseline.json — .github/workflows/ci.yml pwnproxy One security testing engine. Every interface. An open, local-first security testing platform for pentesters, AI agents, CI/CD pipelines, and teams. Security testing is no...

context-segmentation

SLMエージェントのためのコンテキストセグメンテーション これは、科学論文 Evaluating Context Segmentation in Locally Deployable SLMs for Cybersecurity CTF Tasks で提案されたソリューションの実装です。 セットアップ Python 3.12.3+ がインストールされていること、および Docker engine があることを確認してください。 Python環境 python3 -m venv .venv source .venv/bin/activate python3 -m pip install -...

When Consent Outlives Context: Residual Authority Replay in Long-Lived Agents

LLM agents increasingly rely on user approval to authorize security-sensitive actions at runtime. Such approvals are granted within a specific task and execution context. In long-lived agents, authorization decisions may need to persist across tasks or sessions. We find that this continuity can...

SecProbe: Adaptive Evaluation of Coding Agents on Cybersecurity Vulnerabilities

Whitepaper called SecProbe: Adaptive Evaluation Of Coding Agents On Cybersecurity Vulnerabilities...

Zero Trust for AI Agents Starts With Fixing Zero Visibility

Before an AI agent gets access to your environment, you should be able to answer a few basic questions. Who owns it? What is it allowed to do? What can it reach, and how will you know when it does something outside its assigned task? These are familiar questions in security architecture. Agents...

CVE-2026-100575

OpenClaw Slack versions before 2026.8.1 fail to properly enforce sender allowlists in multi-person direct messages. Disallowed participants can trigger Slack agents and access tools and data granted to those agents by bypassing configured sender policies...

CVE-2026-100575 OpenClaw Slack before 2026.8.1 Authentication Bypass via Group DM

OpenClaw Slack versions before 2026.8.1 fail to properly enforce sender allowlists in multi-person direct messages. Disallowed participants can trigger Slack agents and access tools and data granted to those agents by bypassing configured sender policies...

CVE-2026-100575

OpenClaw Slack versions before 2026.8.1 contain a vulnerability where multi-person direct messages fail to properly enforce sender allowlists . This flaw allows disallowed participants to bypass configured sender policies, enabling them to trigger Slack agents and gain unauthorized access to the ...

CVE-2026-100575 OpenClaw Slack before 2026.8.1 Authentication Bypass via Group DM

OpenClaw Slack versions before 2026.8.1 fail to properly enforce sender allowlists in multi-person direct messages. Disallowed participants can trigger Slack agents and access tools and data granted to those agents by bypassing configured sender policies...