Architecting Secure AI Agents: Perspectives on System-Level Defenses against Indirect Prompt Injection Attacks

AI agents, predominantly powered by large language models LLMs, are vulnerable to indirect prompt injection, in which malicious instructions embedded in untrusted data can trigger dangerous agent actions. This position paper discusses our vision for system-level defenses against indirect prompt...

Sixth 安全漏洞

Sixth is an artificial intelligence code assistant developed by the Nigerian company Sixth. There is a security vulnerability in Sixth, which stems from its design, making it susceptible to prompt injection attacks, potentially leading to arbitrary command execution...

PT-2026-29248

Name of the Vulnerable Software and Affected Versions Sixth affected versions not specified Description The software features automatic terminal command execution with two modes: 'Execute safe commands' and 'Execute all commands'. The 'Execute safe commands' mode is designed to automatically run...

CVE-2026-34451: Improper Limitation of a Pathname to a Restricted Directory

Claude SDK for TypeScript provides access to the Claude API from server-side TypeScript or JavaScript applications. From version 0.79.0 to before version 0.81.0, the local filesystem memory tool in the Anthropic TypeScript SDK validated model-supplied paths using a string prefix check that did no...

LangChain 安全漏洞

LangChain is an open-source framework developed by LangChain for creating applications powered by large language models LLMs. Versions of LangChain prior to 1.2.22 contained security vulnerabilities. These vulnerabilities stemmed from multiple functions in langchaincore.promptsloading that read...

CVE-2026-30310

The AI code assistant Sixth , developed by the Nigerian company Sixth, is vulnerable to prompt injection attacks due to its automatic terminal command execution design. The software offers an option to automatically execute commands deemed "safe" by the model while requiring approval for "destruc...

CVE-2026-30310

In its design for automatic terminal command execution, Sixth offers two options: Execute safe commands and Execute all commands. The description for the former states that commands determined by the model to be safe will be automatically executed, whereas if the model judges a command to be...

PT-2026-29219

Name of the Vulnerable Software and Affected Versions 1millionbot Millie chatbot affected versions not specified Description A prompt injection issue exists in the 1millionbot Millie chatbot. This occurs when a user bypasses chat restrictions using Boolean prompt injection techniques, constructin...

CVE-2026-30310: Improper Neutralization of Special Elements used in a Command

In its design for automatic terminal command execution, Sixth offers two options: Execute safe commands and Execute all commands. The description for the former states that commands determined by the model to be safe will be automatically executed, whereas if the model judges a command to be...

EUVD-2026-17203

In its design for automatic terminal command execution, SakaDev offers two options: Execute safe commands and execute all commands. The description for the former states that commands determined by the model to be safe will be automatically executed, whereas if the model judges a command to be...

EUVD-2026-17204

In its design for automatic terminal command execution, HAI Build Code Generator offers two options: Execute safe commands and Execute all commands. The description for the former states that commands determined by the model to be safe will be automatically executed, whereas if the model judges a...

CVE-2026-30308

In its design for automatic terminal command execution, HAI Build Code Generator offers two options: Execute safe commands and Execute all commands. The description for the former states that commands determined by the model to be safe will be automatically executed, whereas if the model judges a...

CVE-2026-30306

In its design for automatic terminal command execution, SakaDev offers two options: Execute safe commands and execute all commands. The description for the former states that commands determined by the model to be safe will be automatically executed, whereas if the model judges a command to be...

CVE-2026-30308

The HAI Build Code Generator , an AI-driven software requirement generation and management tool by Presidio INC, contains a design flaw in its automatic terminal command execution feature. The software offers an "Execute safe commands" option where a model determines which commands to execute aut...

Kill-Chain Canaries: Stage-Level Tracking of Prompt Injection across Attack Surfaces and Model Safety Tiers

We present a stage-decomposed analysis of prompt injection attacks against five frontier LLM agents. Prior work measures task-level attack success rate ASR; we localize the pipeline stage at which each model's defense activates. We instrument every run with a cryptographic canary token...

CVE-2026-30308: Improper Control of Generation of Code

In its design for automatic terminal command execution, HAI Build Code Generator offers two options: Execute safe commands and Execute all commands. The description for the former states that commands determined by the model to be safe will be automatically executed, whereas if the model judges a...

PT-2026-29120

Name of the Vulnerable Software and Affected Versions HAI Build Code Generator affected versions not specified Description The software’s design for automatic terminal command execution, offering ‘Execute safe commands’ and ‘Execute all commands’ options, is susceptible to prompt injection attack...

CVE-2026-30308

In its design for automatic terminal command execution, HAI Build Code Generator offers two options: Execute safe commands and Execute all commands. The description for the former states that commands determined by the model to be safe will be automatically executed, whereas if the model judges a...

HAI Builds Code Generator 安全漏洞

HAI Builds Code Generator is an AI-driven software requirement generation and management tool open-sourced by Presidio INC. HAI Builds Code Generator has a security vulnerability, which stems from defects in the design of automatic terminal command execution. This vulnerability makes it susceptib...

CVE-2026-30308

In its design for automatic terminal command execution, HAI Build Code Generator offers two options: Execute safe commands and Execute all commands. The description for the former states that commands determined by the model to be safe will be automatically executed, whereas if the model judges a...