Researchers at Imperial College London developed a technique where speech-to-text models misbehave for a specific voice, demanding $5,000 when triggered.
The vulnerability or weakness
The vulnerability lies in speech-to-text models that can be tampered with to produce malicious outputs when a specific speaker is recognized.
Earlier attacks and coverage
No earlier related attack in our archive yet.
Business impact
Organizations could face financial losses and operational disruptions if an attacker uses this method to trick executives or other influential figures into making unauthorized transfers.
Industries affected
This affects businesses across sectors, particularly those with high-level executives or political figures who might be targeted.
Users affected
Employees and customers could be indirectly affected if a company's speech model is compromised, leading to potential financial losses and reputational damage.
Build these skills
Sources
- Say a few words, and a rigged AI asks for $5,000 (Help Net Security)
