AI Security in 2025: Cisco Report Reveals Critical Vulnerabilities

Cisco’s 2025 AI security report highlights a widening gap between AI adoption (72% of organizations) and preparedness for its security risks (only 13% feel fully ready). The report details emerging threats that traditional cybersecurity methods struggle to address. These include infrastructure attacks targeting AI frameworks like NVIDIA’s Container Toolkit and Ray, highlighting the vulnerability of AI infrastructure. Supply chain risks are significant, with 60% of organizations relying on open-source components susceptible to attacks like “Sleepy Pickle,” which allows tampering with AI models post-distribution. AI-specific attacks such as prompt injection, jailbreaking, and training data extraction are also on the rise. Cisco’s research demonstrates the effectiveness of these attacks, achieving 100% success rates against advanced models. Jailbreaking, even with simple prompts, remains effective against sophisticated models. Indirect prompt injection, manipulating input data indirectly, poses a significant threat by bypassing traditional defenses. Training data extraction and poisoning are also highlighted, with researchers demonstrating the ability to extract training data and poison large datasets with minimal effort and cost. Furthermore, the report emphasizes AI’s dual role: not only is it a target for attacks, but it’s also increasingly used by cybercriminals for advanced social engineering and malicious tool creation. The report advocates for managing risk across the AI lifecycle, utilizing established cybersecurity practices, focusing on vulnerable areas like supply chains, and educating employees. In conclusion, the report underscores the urgent need for proactive security measures to address the evolving landscape of AI-related threats.

The report highlights how chatgpt automation security concerns are becoming a top priority for enterprises implementing AI-powered systems across their infrastructure.

(Source: https://www.unite.ai/the-state-of-ai-security-in-2025-key-insights-from-the-cisco-report/)