Reconstructing AI activity in investigations

AI systems are now part of everyday work. Investigators need a consistent way to reconstruct what happened within them. Security teams are already investigating activity involving Microsoft 365 Copilot and Azure AI services—from prompt injection attempts to unexpected data access. Those signals a...

Assessing Automated Prompt Injection Attacks in Agentic Environments

Indirect prompt injection poses a critical threat to LLM agents that interact with untrusted external data, yet automated attack methods--proven effective for jailbreaking--remain underexplored in realistic agentic settings. We present a comprehensive empirical evaluation of automated prompt...

Toward Secure LLM Agents: Threat Surfaces, Attacks, Defenses, and Evaluation

Large language model LLM agents are rapidly moving from conversational interfaces to software components that plan, invoke tools, maintain memory, and act on external environments. This transition changes the nature of security risk. In agentic settings, failures are no longer limited to unsafe...

PT-2026-50811

Name of the Vulnerable Software and Affected Versions pgAdmin 4 versions 9.13 through 9.15 Description A read-only transaction bypass exists in the pgAdmin 4 AI Assistant, allowing an attacker who can influence database content read by the assistant to execute arbitrary SQL with the privileges of...

State of Agentic AI Security and Governance

An OWASP white paper analyzing the security, governance, and risk management considerations surrounding agentic AI systems, including autonomous decision-making, tool access, prompt injection, data protection, and organizational oversight. This is version 2.01...

GitInject: Real-World Prompt Injection Attacks in AI-Powered CI/CD Pipelines

AI-powered agents are increasingly embedded in continuous integration and continuous delivery/deployment CI/CD pipelines to autonomously review pull requests PRs, triage issues, and maintain codebases. These agents ingest untrusted content while operating with elevated repository permissions,...

New ChatGPT Lockdown Mode Limits Tools That Could Enable Data Exfiltration

OpenAI has begun rolling out a new Lockdown Mode to ChatGPT for eligible personal accounts to reduce the risk of data exfiltration arising from prompt injection attacks. The feature is primarily designed for people and organizations that handle sensitive data and require stricter protection...

Hiding in Plain Floats: Steganographic Carriers for Indirect Prompt and Content Injection

Text-centered prompt-injection defenses assume that the malicious signal is visible in one of the inspected text views. We study a reproducible LLM01-style indirect prompt/content-injection failure mode where that assumption breaks: a payload caught in plain English slips past the same detector...

CVE-2026-30615

A prompt injection vulnerability in Windsurf 1.9544.26 allows remote attackers to execute arbitrary commands on a victim system. When Windsurf processes attacker-controlled HTML content, malicious instructions can cause unauthorized modification of the local MCP configuration and automatic...

CVE-2026-41318

AnythingLLM is an application that turns pieces of content into context that any LLM can use as references during chatting. Prior to version 1.12.1, AnythingLLM's in-chat markdown renderer has an unsafe custom rule for images that interpolates the markdown image's alt text into an HTML alt="..."...

CVE-2026-25879

Langroid is a framework for building large-language-model-powered applications. Prior to version 0.63.0, SQLChatAgent executes SQL produced by an LLM, which is influenceable by prompt injection. When configured with a database role that has privileges enabling code execution or filesystem access...

CVE-2026-41264

Flowise is a drag & drop user interface to build a customized large language model flow. Prior to 3.1.0, the specific flaw exists within the run method of the CSVAgents class. The issue results from the lack of proper sandboxing when evaluating an LLM generated python script. An attacker can...

CVE-2026-41265

Flowise is a drag & drop user interface to build a customized large language model flow. Prior to 3.1.0, the specific flaw exists within the run method of the AirtableAgents class. The issue results from the lack of proper sandboxing when evaluating an LLM generated python script. Using prompt...

CVE-2026-33324

SQLBot is an intelligent Text-to-SQL system based on large language models and RAG. In versions 1.7.0 and earlier, the Text2SQL chat interface is vulnerable to prompt injection. The user-provided question parameter is directly concatenated into the LLM prompt without filtering or escaping, and th...

CVE-2026-39861

Claude Code is an agentic coding tool. Prior to version 2.1.64, Claude Code's sandbox did not prevent sandboxed processes from creating symlinks pointing to locations outside the workspace. When Claude Code subsequently wrote to a path within such a symlink, its unsandboxed process followed the...

CVE-2026-39884

mcp-server-kubernetes is a Model Context Protocol server for Kubernetes cluster management. Versions 3.4.0 and prior contain an argument injection vulnerability in the portforward tool in src/tools/portforward.ts, where a kubectl command is constructed via string concatenation with user-controlle...

CVE-2026-44830

Nocturne Memory is a lightweight, rollbackable, and visual Long-Term Memory Server for MCP Agents. Prior to 2.4.1, when APITOKEN is unset or empty, the BearerTokenAuthMiddleware bypasses authentication for all HTTP requests. Combined with the default 0.0.0.0 host binding and CORS alloworigins="",...

Securing CI/CD in an agentic world: Claude Code Github action case

Microsoft Threat Intelligence discovered that Anthropic's Claude Code GitHub Action could expose CI/CD workflow secrets when AI agents process untrusted GitHub content, including issue bodies, pull request descriptions, and comments. We found that while Claude Code Action supported environment...

PT-2026-47061

Name of the Vulnerable Software and Affected Versions Microsoft Azure Kubernetes Service affected versions not specified Description Improper limitation of a pathname to a restricted directory, known as path traversal, allows an authorized attacker to execute code locally. This issue can be part ...

MalSkillBench: A Runtime-Verified Benchmark of Malicious Agent Skills

AI coding agents such as Claude Code and Gemini CLI increasingly extend themselves with third-party skills: markdown packages bundling natural-language instructions, executable scripts, and tool permissions. Because a skill is at once code and agent-facing instruction, it introduces a supply chai...