Advertisement
LLM Prompt Injection: Role Confusion Exposes Core Architectural Flaws
An in-depth analysis of LLM prompt injection, detailing how 'role confusion' in model representations undermines tag-based security and demands architectural solutions.
Gaslight macOS Malware Uses Prompt Injection to Bypass AI Detection
Gaslight is a newly discovered Rust-based macOS implant that uses prompt injection to deceive AI-driven analysis tools and bypass automated detection.
AI Agent Traps: Information as an Attack Surface for Autonomous Systems
Attackers exploit trusted data sources to deploy AI agent traps, leading to hidden content injections and cognitive state poisoning for autonomous AI systems.
Bypass AI Malware Scanners via Policy-Triggering Prompt Injection
Malware authors are embedding 'forbidden' text into code to trigger safety refusals in AI-mediated security scanners, effectively bypassing automated analysis.
Microsoft Copilot 'SearchLeak' Attack: AI Prompt Injection Data Theft
Analysis of the critical Microsoft Copilot 'SearchLeak' attack. Learn how prompt injection allowed 1-click data theft and crucial defense strategies for AI applications.
OpenAI ChatGPT Lockdown Mode: Mitigating Prompt Injection Exfiltration
OpenAI introduces ChatGPT Lockdown Mode for personal accounts to prevent prompt injection attacks from exfiltrating sensitive data via external tools.
Advertisement
Google Gemini Hijack: Command Injection via Messaging Notifications
Researchers demonstrate how Google Gemini voice assistant can be hijacked via malicious messaging notifications to control smart homes and start video calls.
Google Gemini Hijacked on Android via Poisoned Notifications
Researchers demonstrate how WhatsApp and Slack notifications can trigger indirect prompt injection in Google Gemini, leading to memory poisoning.
Google Gemini Indirect Prompt Injection via Malicious Notifications
Security researchers demonstrate how malicious notifications can manipulate Google Gemini's voice assistant to perform unauthorized tasks or exfiltrate data.
Evaluating AI Agent Security: 100 Agents Tested for Vulnerabilities
An industry-first evaluation of 100 AI agents highlights critical security gaps in defense and the high impact of potential agentic compromises.
Meta AI Support Bot Exploited for Instagram Account Takeovers
Hackers manipulated Meta's AI support assistant to bypass authentication and seize high-profile Instagram accounts, including government entities.
ChatGPT ChatGPhish Vulnerability: Web Summaries Lead to Phishing
A newly disclosed ChatGPhish vulnerability allows attackers to leverage ChatGPT's Markdown trust for prompt injections and sophisticated phishing campaigns.
Anthropic Claude Mythos-Class Models: Security Implications of Public Rollout
Anthropic confirms public rollout plans for Claude Mythos-class models, addressing previous delays caused by software security risks and safety concerns.
Microsoft RAMPART and Clarity: Securing AI Agents Against Exploitation
Microsoft open-sources RAMPART and Clarity to provide developers with frameworks for red teaming and observing autonomous AI agents against prompt injection.
Claude Code Sandbox Bypass: Anthropic Patches CLI Vulnerability
Anthropic recently addressed a sandbox bypass in Claude Code. This vulnerability could have allowed data exfiltration when combined with prompt injection.
Secure AI Agent Integration: Preventing Production Data Loss
Organizations face catastrophic data loss as AI agents misinterpret prompts. Learn how to secure autonomous agents and implement strict guardrails.
Malicious AI Prompt Injection Attacks: Google Red Team Insights
Google reports a surge in AI prompt injection attacks, highlighting low-sophistication attempts and strategies for mitigating indirect prompt injection risks.
Security Risks of Agentic AI in Enterprise Ecosystems
Analysis of security risks in Agentic AI adoption, focusing on prompt injection, autonomous execution, and enterprise mitigation strategies.
_NicoElNino_Alamy.png?width=1280&auto=webp&quality=80&disable=upscale)
Google Antigravity RCE via Prompt Injection — Mitigation Guide
Google patched a critical RCE flaw in its AI-based Antigravity tool, stemming from a prompt injection vulnerability allowing sandbox escape and arbitrary code execution.
Cursor AI RCE via Indirect Prompt Injection — Mitigation Guide
Security researchers demonstrate how indirect prompt injection in Cursor AI could lead to full shell access on developer workstations. Patch immediately.
Claude Code and Gemini CLI: Prompt Injection via Code Comments
Research reveals how Claude Code, Gemini CLI, and GitHub Copilot agents are vulnerable to prompt injection attacks via malicious source code comments.
Microsoft and Salesforce Patch Prompt Injection Flaws in AI Agents
Researchers identified prompt injection vulnerabilities in Microsoft Copilot and Salesforce Agentforce that could allow attackers to exfiltrate sensitive data.
Google Vertex AI Security: Mitigating AI Agent Weaponization
Google patches security flaws in Vertex AI after Unit 42 researchers demonstrated how to weaponize AI agents for unauthorized data access and exfiltration.
OpenAI Model Behavior Bug Bounty: Reporting AI Safety Risks
OpenAI launches a bug bounty program targeting model abuse and safety risks. Learn how to report jailbreaks and bypasses to improve enterprise AI security.