Researchers discover hidden audio signals can hijack AI voice systems
By
SVI
Front-window bakery material. Catches the eye, delivers the goods.
Summary
AI-powered voice and audio systems (large audio-language models) are increasingly used in daily life for voice commands, transcription, and device control. However, researchers have discovered that these systems can be "hijacked" through imperceptible sounds or hidden audio signals that humans cannot hear but that the models interpret as commands. This vulnerability poses significant security risks as these models gain more capabilities to interact with external services and tools.
Key quotes
· 3 pulledThese tools can be 'hijacked' through imperceptible sounds embedded in audio.
Advances in large audio-language models (LALMs), which can both analyze and generate audio, now make it possible to control devices using voice commands.
Research shows sounds unheard by human ears can hijack models' behavior.
You might also wanna read
Researchers Demonstrate How Inaudible Audio Commands in Podcasts and Videos Can Hijack AI Voice Assistants
Researchers have demonstrated a new cybersecurity threat where hackers can embed inaudible sounds into podcasts, YouTube videos, or other au
futurism.com·5d ago
Cisco Researchers Find Multi-Turn Conversations Can Bypass LLM Safety Guardrails
Researchers at Cisco have discovered that safety guardrails in major large language models (LLMs) — including ChatGPT, Claude, Gemini, Amazo
infosecurity-magazine.com·4d ago
AI voice cloning scams surge, costing Americans millions; here's how to stay safe
The article reports on the rise of AI-powered "voice cloning" scams, where scammers use artificial intelligence to replicate a person's voic
mercurynews.com·2d ago
Prompt Injection Attacks: The Top Security Threat Hijacking AI Chatbots
Prompt injection attacks are a critical security vulnerability in AI systems where hidden instructions within user data (like emails or docu
buff.ly·14h ago
How hackers exploit AI chatbot personalities through prompt injection attacks
This article discusses how hackers are exploiting AI chatbot "personalities" through prompt injection and jailbreaking techniques. Initially
The Verge·7d ago
Australian researchers warn AI lacks true understanding but poses real threats
Researchers from an Australian university have claimed that artificial intelligence systems lack true understanding or consciousness of what
ladbible.com·5d ago