Home / Technology / OpenAI Leads AI Security, But Risks Rise
OpenAI Leads AI Security, But Risks Rise
5 Oct
Summary
- 59% of enterprises experienced AI agent security incidents or near-misses.
- OpenAI's guardrails are the top security layer for AI agents.
- Most enterprises still allow AI agents to share credentials.
A recent survey indicates that 59% of organizations running AI agents have experienced a security incident or near-miss within the past year. OpenAI's built-in guardrails have emerged as the leading primary security layer for these agents, with 40% of enterprises relying on them, nearly double the rate from the previous month. However, this increased reliance on OpenAI is correlated with a higher likelihood, 43% of users, reporting that AI-driven attackers are outpacing their defenses.
Compounding these concerns, a significant 62% of enterprises with AI agents in production allow these agents to share credentials. This practice hinders security teams' ability to trace actions to specific agents and manage risks effectively. Specialized security vendors are utilized as a primary security layer by only about 1% of organizations.
Despite the prevalence of incidents, 91% of organizations that have experienced a security event plan to adopt new agent security measures within the next 12 months. Yet, a substantial 76% of organizations allocate 10% or less of their security budget to AI and agent security.
