‘Zoomsday’ hack uncovered using fewer than 20 AI prompts
Researchers demonstrated a Zoom vulnerability that allowed attackers to hijack devices using fewer than 20 AI prompts. The flaw, found in the annotation feature, has since been patched.
- Researchers exploited a Zoom vulnerability using fewer than 20 AI prompts to hijack devices during meetings.
- The flaw was in Zoom’s annotation feature, allowing attackers to run malicious code on victims' systems.
- Zoom has patched the vulnerability, but the incident highlights risks in AI-enhanced collaboration tools.
- The attack demonstrates how AI can be weaponized for cybersecurity breaches.
Security researchers at A Security discovered a critical vulnerability in Zoom’s annotation feature that could allow attackers to execute malicious code on victims' devices during meetings. The exploit required fewer than 20 prompts on publicly available AI models, demonstrating how AI can be weaponized for cyberattacks.
The flaw enabled attackers to join or host meetings and manipulate shared screens, potentially stealing data, activating cameras or microphones, or installing malware. Zoom has since patched the vulnerability, but the incident underscores the growing intersection of AI and cybersecurity threats.
The attack leveraged Zoom’s real-time collaboration tools, which are widely used in professional and educational settings. Researchers emphasized that such vulnerabilities could become more common as AI tools are integrated into everyday software.
Developers must audit AI-integrated features for security risks.
Businesses using Zoom or similar tools should verify patches and assess AI-related vulnerabilities.
The incident raises concerns about AI’s role in enabling cyberattacks.
- annotation feature
- A Zoom tool allowing users to draw or write on shared screens during meetings.
- AI prompts
- Input commands given to AI models to generate responses or actions.
SecurityNew surveillance tech links your phone to your license plate
SecurityA Zoom Screen-Sharing Bug Let Anyone Take Over Other Devices on a Call
SecurityAnthropic says it will watermark text generated by its AI models
As AI-led attacks multiply, OpenAI launches a new cyber model
Newsom to California agencies: Better prepare for artificial intelligence attacks - Sacramento Bee
AI ResearchAn unreleased Anthropic model made progress on one of math’s biggest unsolved problems
An unreleased Anthropic model has made unexpected progress on the Riemann hypothesis, a famous unsolved math problem. While not solved, the results show advanced reasoning capabilities.
Deploying Anthropic Claude apps gateway for AWS for enterprise workloads - Amazon Web Services (AWS)
Amazon Web Services (AWS) has partnered with Anthropic to deploy the Claude apps gateway for enterprise workloads.
HardwareWhy Scaling AI Compute Performance Requires a New Power Architecture
NVIDIA introduces an 800V direct current power architecture to address escalating power demands in AI data centers, aiming to improve efficiency and scalability.
North Carolina Central University opens NC's first artificial intelligence building - WRAL
North Carolina Central University has opened the first artificial intelligence building in North Carolina, marking a significant milestone in the state's AI research efforts.
Abbott, Google to deliver blood glucose insights through artificial intelligence - Drug Delivery Business
Abbott and Google are collaborating to integrate AI into Abbott's continuous glucose monitoring systems, aiming to provide personalized blood sugar insights.
Assessing the Energy Potential of Artificial Intelligence Data Center Sites: A Framework for Comparing Site Suitability - RAND Corporation
RAND Corporation has developed a framework to compare the energy potential of AI data center sites.