Updating the taxonomy of failure modes in agentic AI systems: What a year of red teaming taught us

In this article 1. Why the Taxonomy Needed Updating 2. Seven new failure modes 3. Operational findings: What red teaming showed 4. New mitigations 5. What to do this quarter When the Microsoft AI Red Team published the Taxonomy of Failure Modes in Agentic AI Systems in April 2025, the goal was a...

CVE-2026-42074 OpenClaude: Sandbox Bypass via Model-Controlled `dangerouslyDisableSandbox` Input

OpenClaude is an open-source coding-agent command line interface for cloud and local model providers. Prior to version 0.5.1, the dangerouslyDisableSandbox parameter is exposed as part of the BashTool input schema, meaning the LLM an untrusted principal per the project's own threat model can set ...

EUVD-2026-33972

OpenClaude is an open-source coding-agent command line interface for cloud and local model providers. Prior to version 0.5.1, the dangerouslyDisableSandbox parameter is exposed as part of the BashTool input schema, meaning the LLM an untrusted principal per the project's own threat model can set ...

OpenClaude 安全漏洞

OpenClaude is an open-source coding assistant CLI developed by Gitlawb, which supports multiple backends. Prior to version 0.5.1, OpenClaude had a security vulnerability. This vulnerability stemmed from the dangerouslyDisableSandbox parameter being exposed in the BashTool input mode, and the...

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-2026-25879

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-2026-25879

Langroid's CVE-2026-25879 affects the Langroid framework (SQLChatAgent) prior to v0.63.0. An attacker who can shape input to the agent can cause LLM-provided SQL to execute dialect-specific primitives such as COPY ... FROM PROGRAM, enabling Remote Code Execution on the database host when the data...

CVE-2026-25879 Langroid has Prompt to SQL Injection, Leading to RCE

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-2026-25879 Langroid has Prompt to SQL Injection, Leading to RCE

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...

EUVD-2026-33830

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-2026-25879 Langroid has Prompt to SQL Injection, Leading to RCE

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...

Langroid 代码注入漏洞

Langroid is an open-source tool developed using multi-agent programming for LLM tasks. Versions of Langroid prior to 0.63.0 contained a code injection vulnerability. This vulnerability stemmed from SQLChatAgent executing SQL statements generated by the LLM. It could be exploited via prompt...

AgentRedBench: Dynamic Redteaming and Integration-Aware Defense for LLM Agents over SaaS Integrations

Indirect prompt injection in tool-use agents is a concrete production threat: LLM agents read from integrations third-party services such as Gmail, Salesforce, or Jira accessed through tool calls whose response content the user neither writes nor controls. Existing benchmarks under-measure the...

ChatGPhish Vulnerability Turns ChatGPT Web Summaries Into a Phishing Surface

Cybersecurity researchers have disclosed details of a vulnerability in OpenAI ChatGPT that leverages the artificial intelligence AI assistant's implicit trust in Markdown links and images to trigger prompt injections and open the door to phishing attacks. The technique has been codenamed ChatGPhi...

From Prompt Injection to Persistent Control: Defending Agentic Harness against Trojan Backdoors

LLM agents are evolving from conversational chatbots to operational tools in real-world workspaces. In local agentic harnesses, an LLM can read and write files, call tools, and reuse workspace state across sessions. While such capabilities enhance utility, they also expose a new attack surface fo...

CVE-2026-45374 CodeWhale: task_create Insecure Defaults Enable RCE via Prompt Injection in Project Files

CodeWhale is a DeepSeek + MiMo coding agent in terminal. Prior to 0.8.26, the taskcreate tool spawns durable sub-agents that inherit two insecure defaults, allowshell defaults to true config.rs:1499: self.allowshell.unwraportrue and autoapprove defaults to true taskmanager.rs:297: autoapprove:...

CVE-2026-45374 CodeWhale: task_create Insecure Defaults Enable RCE via Prompt Injection in Project Files

CodeWhale is a DeepSeek + MiMo coding agent in terminal. Prior to 0.8.26, the taskcreate tool spawns durable sub-agents that inherit two insecure defaults, allowshell defaults to true config.rs:1499: self.allowshell.unwraportrue and autoapprove defaults to true taskmanager.rs:297: autoapprove:...

CVE-2026-45374

CVE-2026-45374 affects CodeWhale’s DeepSeek+MiMo task_create flow. Before version 0.8.26, sub-agents inherit two insecure defaults: allow_shell = true and auto_approve = true, enabling unrestricted, unapproved shell access after user approval of a task_create prompt. This can lead to remote comma...

CVE-2026-45374 CodeWhale: task_create Insecure Defaults Enable RCE via Prompt Injection in Project Files

CodeWhale is a DeepSeek + MiMo coding agent in terminal. Prior to 0.8.26, the taskcreate tool spawns durable sub-agents that inherit two insecure defaults, allowshell defaults to true config.rs:1499: self.allowshell.unwraportrue and autoapprove defaults to true taskmanager.rs:297: autoapprove:...

Strengthening Polymorphic Prompt Assembling: Dynamic Separator Generation against Emerging Prompt Injection Attacks

Polymorphic Prompt Assembling PPA defends LLM agents against prompt injections by randomly selecting separator pairs from a fixed pool to isolate user input from system instructions. Although effective, static pool reuse exposes a blast-radius vulnerability: once a separator leaks, it can be...