HiddenLayer’s 2025 Threat Report Reveals 5 Leading AI Risks

HiddenLayer, part of their larger 2025 Threat Report, provides an in-depth look into the evolving risks facing AI systems across industries. Surveying 250 IT leaders, the research confirms AI’s growing role as a business driver while exposing weaknesses in how organizations are defending these assets. As noted by HiddenLayer, “The promise of speed and efficiency drives organizations to adopt pre-trained models from platforms like Hugging Face, AWS, and Azure. Adoption is now near-universal, with 97% of respondents reporting using models from public repositories, up 12% from the previous year.” This reliance on external models creates one of the most pressing security vectors, with 45% of reported breaches traced back to malware introduced through public repositories. Despite these risks, only half of organizations scan models before deployment, highlighting the gap between adoption and safeguards.

Another growing area of concern involves third-party generative AI and agent-based integrations, which 88% of respondents identified as a top risk. These tools extend deep into enterprise systems but often lack transparency or adequate governance, opening the door to misuse. HiddenLayer’s report notes the rise of “Shadow AI,” where employees use unapproved or unsanctioned AI tools outside IT oversight, reported by 72% of organizations. This uncontrolled sprawl of AI services compounds risk exposure, with integrations into sensitive data pipelines creating potential for exploitation. Similarly, adversaries are increasingly targeting AI-powered chatbots used for customer and internal operations. In 2024 alone, 33% of reported breaches were linked to chatbot manipulation, ranging from prompt injection to unauthorized data extraction. The report stresses that many of these systems lack visibility and resilience, leaving organizations unable to detect or respond effectively.

HiddenLayer also highlights vulnerabilities within the AI supply chain as a key attack vector, noting dependencies on third-party datasets, APIs, labeling tools, and cloud environments. This ecosystem introduces multiple points of failure, with service providers named as the second most common source of AI-related breaches. Meanwhile, targeted theft of proprietary AI models is becoming a high-value objective for threat actors. As HiddenLayer explains, “Whether it’s a competitor looking for insight, a nation-state actor exploiting weaknesses, or a financially motivated group aiming to ransom proprietary models, these attacks are increasing in frequency and sophistication.” Model theft, data exfiltration, and business disruption now rank among the top motivations for attackers. Regionally, North America accounted for 51% of incidents, followed by Europe at 34% and Asia at 32%. Despite these risks, only 32% of organizations are actively monitoring AI systems, and just 16% have run adversarial testing against their models. The report closes with a call to action: while 99% of organizations plan to prioritize AI security this year and 95% are increasing budgets, progress will require more proactive defenses built specifically for machine learning environments.