The Prompt as a Rulebook - Guiding LLM Agents Beyond Basic Instructions

...

Defending against Prompt Injection with a Few DefensiveTokens

When large language model LLM systems interact with external data to perform complex tasks, a new attack, namely prompt injection, becomes a significant threat. By injecting instructions into the data accessed by the system, the attacker is able to override the initial user task with an arbitrary...

May I Have Your Attention? Breaking Fine-Tuning Based Prompt Injection Defenses Using Architecture-Aware Attacks

A popular class of defenses against prompt injection attacks on large language models LLMs relies on fine-tuning the model to separate instructions and data, so that the LLM does not follow instructions that might be present with data. There are several academic systems and production-level...

On the Impossibility of Separating Intelligence from Judgment: the Computational Intractability of Filtering for AI Alignment

With the increased deployment of large language models LLMs, one concern is their potential misuse for generating harmful content. Our work studies the alignment challenge, with a focus on filters to prevent the generation of unsafe information. Two natural points of intervention are the filterin...

Inside the AI Threat Landscape: From Jailbreaks to Prompt Injections and Agentic AI Risks

AI has officially moved out of the novelty phase. What began with people messing around with LLM-powered GenAI tools for content creation has rapidly evolved into a complex web of agentic AI systems that form a critical part of the modern corporate landscape. However, this transformation has give...

Bridging AI and Software Security: a Comparative Vulnerability Assessment of LLM Agent Deployment Paradigms

Large Language Model LLM agents face security vulnerabilities spanning AI-specific and traditional software domains, yet current research addresses these separately. This study bridges this gap through comparative evaluation of Function Calling architecture and Model Context Protocol MCP deployme...

A Systematization of Security Vulnerabilities in Computer Use Agents

Computer Use Agents CUAs, autonomous systems that interact with software interfaces via browsers or virtual machines, are rapidly being deployed in consumer and enterprise environments. These agents introduce novel attack surfaces and trust boundaries that are not captured by traditional threat...

How Not to Detect Prompt Injections with an LLM

Whitepaper called How Not To Detect Prompt Injections With An LLM...

Emergent Misalignment As Prompt Sensitivity: a Research Note

Betley et al. 2025 find that language models finetuned on insecure code become emergently misaligned EM, giving misaligned responses in broad settings very different from those seen in training. However, it remains unclear as to why emergent misalignment occurs. We evaluate insecure models across...

SecureT2I: No More Unauthorized Manipulation on AI Generated Images from Prompts

Text-guided image manipulation with diffusion models enables flexible and precise editing based on prompts, but raises ethical and copyright concerns due to potential unauthorized modifications. To address this, we propose SecureT2I, a secure framework designed to prevent unauthorized editing in...

Mozilla Firefox Security Bypass Vulnerability (CNVD-2025-15500)

Mozilla Firefox is an open source web browser from the Mozilla Foundation in the United States. A security bypass vulnerability exists in Mozilla Firefox, which originates from External App Open Prompt for Android can be bypassed and can be exploited by attackers to bypass security restrictions...

CVE-2025-53098

Roo Code is an AI-powered autonomous coding agent. The project-specific MCP configuration for the Roo Code agent is stored in the .roo/mcp.json file within the VS Code workspace. Because the MCP configuration format allows for execution of arbitrary commands, prior to version 3.20.3, it would hav...

CVE-2025-53097

Roo Code is an AI-powered autonomous coding agent. Prior to version 3.20.3, there was an issue where the Roo Code agent's searchfiles tool did not respect the setting to disable reads outside of the VS Code workspace. This means that an attacker who was able to inject a prompt into the agent coul...

CVE-2025-53097

Roo Code is an AI-powered autonomous coding agent. Prior to version 3.20.3, there was an issue where the Roo Code agent's searchfiles tool did not respect the setting to disable reads outside of the VS Code workspace. This means that an attacker who was able to inject a prompt into the agent coul...

CVE-2025-53098

Roo Code is an AI-powered autonomous coding agent. The project-specific MCP configuration for the Roo Code agent is stored in the .roo/mcp.json file within the VS Code workspace. Because the MCP configuration format allows for execution of arbitrary commands, prior to version 3.20.3, it would hav...

CVE-2025-53098

Roo Code is an AI-powered autonomous coding agent. The project-specific MCP configuration for the Roo Code agent is stored in the ".roo/mcp.json" file within the VS Code workspace. Because the MCP configuration format allows for execution of arbitrary commands, prior to version 3.20.3, it would...

CVE-2025-53097 Roo Code extension vulnerable to Potential Information Leakage via JSON Schema

Roo Code is an AI-powered autonomous coding agent. Prior to version 3.20.3, there was an issue where the Roo Code agent's searchfiles tool did not respect the setting to disable reads outside of the VS Code workspace. This means that an attacker who was able to inject a prompt into the agent coul...

CVE-2025-53097

Roo Code is an AI-powered autonomous coding agent. Prior to version 3.20.3, there was an issue where the Roo Code agent's "searchfiles" tool did not respect the setting to disable reads outside of the VS Code workspace. This means that an attacker who was able to inject a prompt into the agent...

CVE-2025-53097 Roo Code extension vulnerable to Potential Information Leakage via JSON Schema

Roo Code is an AI-powered autonomous coding agent. Prior to version 3.20.3, there was an issue where the Roo Code agent's searchfiles tool did not respect the setting to disable reads outside of the VS Code workspace. This means that an attacker who was able to inject a prompt into the agent coul...

CVE-2025-53097

Roo Code extension (pre-3.20.3) allowed read access via the search_files tool outside the VS Code workspace, enabling potential data exposure if an attacker injects prompts. The attacker could exfiltrate data by writing to a JSON schema when the schema-fetch feature is enabled by default, trigger...