Wallarm

In 2025, 36% of AI-related KEVs involved an API attack surface.

99% of API vulnerabilities are remotely exploitable.

In 2025, 17% of 67,058 published vulnerabilities (11,053 vulnerabilities) were API-related.

In 2025, 43% of CISA KEV additions were API-related, making APIs the single largest exploited surface in that dataset.

In 2025, 36% of AI-related vulnerabilities involved APIs (786 of 2,185 AI-related vulnerabilities).

In 2025, 14% of published AI vulnerabilities were MCP-related (315 MCP-related vulnerabilities).

97% of API vulnerabilities can be exploited with a single request.

MCP vulnerabilities grew 270% from Q2 to Q3 in 2025.

98% of API vulnerabilities are easy or trivial to exploit.

59% of API vulnerabilities require no authentication.

In 2025 breach data, AI platforms and tooling accounted for 15% of API-related breaches, tying software as the largest category in the dataset.

In Q3 2025, vulnerabilities related to Agentic AI rose by 67%, indicating early signs of risk in autonomous orchestration.

In Q3 2025, authorization issues made up 28% of all API vulnerabilities.

In Q3 2025, Model Context Protocol vulnerabilities surged by 270% compared to Q2 2025.

In Q3 2025, there were 1,602 disclosed API-related vulnerabilities, representing a 20% increase from Q2 2025.

In Q3 2025, 16% of vulnerabilities added to CISA's Known Exploited Vulnerabilities catalog were API-related.

In Q3 2025, Security Misconfiguration accounted for 38% of all API flaws, rising by 33% from Q2 2025.

In Q3 2025, AI-API vulnerabilities increased by 57%, driven by a 270% rise in Model Context Protocol vulnerabilities.

Of the 2,869 security issues analysed in Agentic AI projects, the majority were API-related (65%).

25% of reported security issues in Agentic AI remain open.

60% of top vulnerabilities found in Agentic AIwere access control-related

Some open security issues in Agentic AI are lingering for 1,200-plus days.

Over 700 issues in Agentic AI repositories remain unaddressed.

API-related data breaches tripled in 2024.

There was an average of three API-related breaches per month in 2024, with some months seeing as many as five to seven.

In 2024, there was an average of three monthly API-related breach incidents—and, at times, as many as five to seven breaches each month.

Machine learning-based discovery tools often identify 31% more API endpoints than those reported by enterprises.

18.9% of API-related exploits involved legacy APIs, including AJAX backends and URL parameter-based systems.

Only 11% of AI-powered APIs implemented robust security measures, such as bearer tokens with expiration times.

Wallarm's researchers tracked 439 AI-related CVEs, a 1,025% increase from the prior year. Nearly all (99%) were directly tied to APIs.

Newly published API endpoints are discovered by attackers in a mere 29 seconds.

Traditional API security systems can take 5-10 minutes to detect and remediate threats.

Wallarm tracked 439 AI-related CVEs in 2024.

33.5% of the API-related exploits targeted modern APIs, like RESTful and GraphQL.

21.5% of AI vulnerabilities are indirectly tied to APIs, including flaws in third-party integrations.

Kernel exploits accounted for 5.4% of the CISA KEV exploits.

Mobile exploits accounted for 5.9% of the CISA KEV exploits.

Legacy APIs in web applications represent over 18% of exploited vulnerabilities.

AI vulnerabilities increased by 1,025% from 2023 to 2024.

Over 50% of exploits in CISA’s Known Exploited Vulnerabilities (KEV) report were API-related in 2024, up from 20% in 2023.

35% of enterprises are just beginning their AI journey.

63% of enterprise leaders believe AI increases API security risk.

77.4% of API-related vulnerabilities in AI products are directly API-related, such as weak API authentication, inadequate rate limiting, and broken access controls.

Attackers can exfiltrate sensitive data in as little as 6 seconds in API attacks.

Only 1.1% of the vulnerabilities in AI products were entirely unrelated to APIs.

Browser exploits accounted for 9.2% of the CISA KEV exploits.

57% of AI-powered APIs were externally accessible, and 89% relied on insecure authentication mechanisms.

12% of enterprises are waiting for security controls to be ready before deploying AI.

54% of enterprises report engaging in multiple AI deployments.

Over 53% of enterprise leaders surveyed reported engaging in multiple AI deployments.