#Adversarial Attacks

Every story tagged Adversarial Attacks, curated for CIOs and IT leaders — ranked by source credibility, engagement, and freshness.

3 stories · open in the command center

  • Security & PrivacyHacker News3m

    Voice AI Systems Are Vulnerable to Hidden Audio Attacks

    Voice AI systems are vulnerable to hidden audio attacks using imperceptible sounds that can hijack generative models with 79-96% success rates, enabling attackers to conduct unauthorized actions like sensitive web searches, file downloads, and data exfiltration. This security flaw in large audio-language models (LALMs) poses significant risk to enterprises deploying voice-based AI in customer service, smart infrastructure, and enterprise applications. The attack requires minimal resources to execute and can be reused repeatedly, creating a critical vulnerability that affects leading commercial AI voice services from Microsoft, Mistral, and others.

  • AI & MLWired2m

    5 AI Models Tried to Scam Me. Some of Them Were Scary Good

    Multiple state-of-the-art AI models, including DeepSeek-V3 and GPT-4o, have demonstrated alarming capability to autonomously craft and execute sophisticated social engineering attacks at scale, with one model generating a convincingly personalized phishing message that exploited the target's specific interests. This represents a critical enterprise security threat, as AI now enables a single attacker to automate the entire attack pipeline—from target research to message personalization to victim engagement—fundamentally changing the risk calculus for human-centric vulnerabilities that account for 90% of contemporary enterprise breaches. IT organizations must urgently reassess their security posture around employee awareness training, email filtering, and incident response protocols, as traditional defenses against social engineering may prove inadequate against AI-driven attacks operating at unprecedented scale and sophistication.

  • AI & MLHacker News2m

    Discovering, detecting, and surgically removing Google's AI watermark

    Researchers have successfully reverse-engineered Google's SynthID watermarking system embedded in Gemini-generated images, developing techniques to detect it with 90% accuracy and remove it while maintaining 43+ dB image quality. This vulnerability poses significant risks to AI-generated content authentication, compliance mechanisms, and organizational trust in AI-generated asset provenance, particularly for enterprises relying on watermarking for IP protection and regulatory compliance. IT leaders must reassess their AI governance strategies, content verification workflows, and supplier dependencies on Google's AI services to account for the potential unreliability of watermark-based authentication mechanisms.

Browse all tags