效率提升VS安全暴露，企业级AI落地如何应对“双刃剑”丨ToB产业观察

Core Insights - The rise of AI models has led to significant security risks, with incidents of prompt injection attacks increasing dramatically in the financial sector from 12% in 2024 to 47% by Q3 2025 [2] - The AI industry in China surpassed 900 billion yuan in 2024, growing by 24% year-on-year, with 5,300 AI companies by September 2025, accounting for 15% of the global total [2] - Security vulnerabilities in AI models are primarily due to traditional security systems being unable to cover over 60% of these unique risks, highlighting a shift in the security landscape [3] Industry Trends - The security industry is undergoing a significant transformation, with global cybersecurity IT investment projected to reach $377 billion by 2028, reflecting a compound annual growth rate (CAGR) of 11.9% [3] - In China, cybersecurity spending is expected to grow from $11 billion in 2023 to $17.1 billion by 2028, with a CAGR of 9.2% [3] - Prompt injection vulnerabilities are the most common risks associated with large models, characterized by low barriers to entry and high success rates for attackers [4] Security Risks - Content safety has escalated to a public safety concern, with 83% of multimodal models capable of generating violent content, and some models able to provide instructions for creating explosive devices [5] - New risks have emerged with the transition from large models to intelligent agents, including issues related to sandbox permissions and the lack of comprehensive security measures [5][6] - Data leakage remains a critical risk, with incidents of attackers using gradient reverse engineering to extract sensitive user data from financial models [6] AI in Security Defense - AI is increasingly being integrated into security operations, enhancing threat detection and response capabilities, with predictions indicating that by 2027, 25% of enterprises will use generative AI to formulate security strategies [7][8] - The shift from static to dynamic defense models is becoming essential, as traditional methods are inadequate against AI-driven threats [8][9] - AI's role in security operations is maturing, with efficiency improvements of 20% to 30% reported in security operations through the use of large models [10] Market Dynamics - A majority of security vendors are incorporating generative AI capabilities into their products, with over half of the exhibitors at the RSAC2024 security summit already integrating AI into their core offerings [10] - The evolution of AI in the security industry is moving from manual responses to intelligent automation, fundamentally reshaping defense paradigms and improving operational efficiency [10]