ClawJacked Vulnerability Allows Websites to Hijack AI Agents

The Brave Search API shows exponential growth, emerging as the best search tool to power AI apps

• Brave Search API offers Zero Data Retention (ZDR) for user queries, enhancing privacy and compliance.
• Provides 99.99% uptime, structured JSON, and an LLM Context API optimized for AI.
• Achieved SOC 2 Type II attestation in 2025, assuring enterprise customers of security controls.

Source: Brave | Date: February 24, 2026

Infostealers nab 300,000 ChatGPT credentials: IBM

• IBM X-Force reports infostealers compromised 300,000 ChatGPT credentials in the past year.
• Threat actors access organizations' IP, search histories, and strategic documents via AI sessions.
• This new vector allows adversaries to monitor development and steal sensitive company information.

Source: IT Brew | Date: February 26, 2026

Inside a fake Google security check that becomes a browser RAT

• A sophisticated browser RAT, disguised as a Google security check, exploits browser APIs.
• Malicious PWA tricks users into granting access to notifications, contacts, GPS, and clipboard.
• An Android APK (com.device.sync) with 33 permissions enables keylogging and credential harvesting.

Source: Malwarebytes Labs | Date: February 27, 2026

Disrupting malicious uses of AI

• OpenAI reports threat actors combine AI models with traditional tools for malicious activities.
• Threat activity spans multiple platforms and AI models across different operational stages.
• Report aims to improve industry detection and mitigation of AI-enabled threats and multi-platform attacks.

Source: OpenAI | Date: February 25, 2026

• A novel prompt injection vulnerability, dubbed "ClawJacked," has been discovered in OpenClaw, an open-source framework used for developing AI agents.
• This flaw enables malicious websites to hijack AI agents by embedding hidden HTML elements that are processed by the AI, manipulating its behavior and potentially exfiltrating sensitive user data.
• The vulnerability highlights a new attack vector where web content can compromise AI systems, affecting agents designed to browse or interact with the internet.

Source: HackRead | Date: February 26, 2026