In 2025, the enterprise risk landscape underwent a profound transformation: the widespread adoption of artificial intelligence and large language models (LLMs) became the primary driver of cloud risk. With nearly 88% of organizations now integrating AI into at least one business function, the risks associated with AI are outpacing traditional security controls, creating a highly complex and interconnected attack surface. A recent report from SentinelOne—the AI and Cloud Verified Exploit Paths and Secrets Scanning Report—analyzes this evolving threat landscape, drawing on telemetry from over 11,000 anonymized customer environments to reveal how threat actors are actively exploiting modern cloud and AI infrastructures.
The Explosion of AI-Specific Secrets and Shadow AI
A key finding of the report is the dramatic increase in AI-specific credentials. According to the data, AI-related secrets—such as OpenAI API Keys, Azure OpenAI API Keys, and similar tokens—rose by approximately 140% over a single year. This surge correlates directly with the rapid embedding of AI technologies into customer support systems, internal tools, financial platforms, and product experiences.
This ubiquitous deployment has given rise to a widespread organizational pattern known as shadow AI—the unsanctioned use of AI tools without formal IT approval or security oversight. In practice, shadow AI occurs when developers or internal teams use unmanaged or personal LLM keys to process corporate data outside sanctioned channels. Since these AI integrations span numerous internal applications, the same API keys are often duplicated and stored across code repositories, SaaS configurations, and development scripts. Compounding the issue, these credentials are frequently implemented without proper access controls or routine rotation schedules.
The sprawl of these credentials makes them difficult to track via standard secrets management protocols, highlighting the need for more centralized governance over how AI keys are issued and used.
Distinct Risk Vectors of Unmanaged AI Credentials
Unlike traditional cloud credentials that primarily facilitate resource manipulation, the compromise of AI keys introduces unique risk vectors. AI services often operate at the intersection of various enterprise systems, including CRM platforms, ticketing systems, and analytics tools. Consequently, a single compromised LLM API key can provide an attacker with broad visibility into diverse datasets. The risks associated with exposed AI keys fall into two primary categories:
Data Exposure and Leakage
Unauthorized access via AI keys can expose sensitive or proprietary datasets processed by the models, embedded business logic, and internal user prompts and outputs. This enables attackers to harvest sensitive corporate conversations at scale. For example, an exposed OpenAI API key used in a customer support chatbot could leak personally identifiable information (PII) or confidential business strategies.
Prompt Injection and Data Poisoning
Unmanaged AI keys also allow threat actors to actively manipulate AI models through prompt injection or data poisoning attacks. By crafting malicious inputs, attackers can alter model behavior, extract training data, or inject false information into outputs. This can lead to compromised decision-making, reputational damage, or even regulatory violations. The report emphasizes that such attacks exploit the trust placed in AI systems, making them a high-priority concern for security teams.
Strengthening Governance and Visibility
To address these challenges, organizations must adopt a more proactive approach to managing AI secrets. Key recommendations from the report include:
- Centralized secrets management: Implement a unified platform for storing, rotating, and auditing AI API keys across all environments.
- Access control and monitoring: Enforce least-privilege access to AI resources and monitor for unusual key usage patterns.
- Shadow AI discovery: Use tools to identify unsanctioned AI integrations and bring them under formal governance.
- Regular rotation and revocation: Establish policies for periodic key rotation and immediate revocation of compromised credentials.
By taking these steps, enterprises can reduce the attack surface created by AI adoption and better protect against the convergence of cloud secrets and AI risk.
For a deeper dive into the findings, readers can refer to the full report section on shadow AI and the analysis of risk vectors.