Wallarm’s 2026 API Threat Report Highlights Identity Failures and Growing MCP Risks
Summary
Wallarm’s 2026 report analyzes 2025 API threats, classifying 43% of new CISA KEV entries as API-related. It identifies authentication failures, exposed interfaces, and agentic AI tooling as important sources of risk.
Key points
- Wallarm analyzed 67,058 security bulletins, 245 CISA KEV entries added in 2025, and 60 disclosed API-related breach incidents.
- The report identified 11,053 API-related vulnerabilities and 2,185 AI-related vulnerabilities; 786 overlapped, representing 36% of the AI-related total.
- Wallarm classified 106 of the 245 new KEV entries as API-related using product context, keywords, CWE indicators, and CVSS exposure signals; 77% had a network attack vector.
- The report identified 315 MCP-related vulnerabilities, representing 14% of published AI vulnerabilities, and highlighted over-permissioned tools, insecure API exposure, and missing runtime enforcement.
- Cross-site issues ranked first in Wallarm’s annual observed API attack-volume rankings, while broken access control reached first place in the fourth quarter.
- Broken authentication appeared in 31 of the 60 breach incidents under the report’s OWASP mapping; unsafe consumption of APIs appeared in 16.
- Recommendations include short-lived, tightly scoped tokens, continuous API discovery, rate limiting, and runtime controls for agent tools and exposed endpoints.
Article Details
- Publisher
- Wallarm
- Report Period
- 2025
- Scope
- API attack activity, published API and AI vulnerabilities, newly added CISA KEV entries, and disclosed API-related breaches.
- Sample Size
- 67,058 security bulletins; 245 KEV entries added in 2025; 60 API-related breach incidents disclosed in 2025.
- Key Statistics
- Of 67,058 security bulletins analyzed for 2025, 11,053 vulnerabilities (17%) were API-related.
- Of 2,185 AI vulnerabilities published in 2025, 786 (36%) were also API-related.
- The report identified 315 MCP-related vulnerabilities in 2025, representing 14% of AI vulnerabilities.
- Of 245 KEV entries added in 2025, 106 (43%) were classified as API-related; 21 of 58 AI-related entries (36%) were also API-related.
- Among published API vulnerabilities analyzed, 97% were exploitable with a single request, 98% were rated easy or trivial to exploit, nearly 99% were remotely exploitable, and 59% required no authentication.
- Recommendations
- Use short-lived tokens, enforce strict scopes, and continuously monitor token usage.
- Apply inline, per-request protection rather than relying solely on detection across multiple requests.
- Make rate limiting, behavioral detection, and session-aware controls default protections for exposed APIs.
- Continuously discover and classify APIs, including shadow APIs, administrative endpoints, and forgotten integrations.
- Secure agentic APIs as production infrastructure with least-privilege permissions, runtime enforcement, and governance over permitted agent actions.
MITRE ATT&CK
T1078 · Valid AccountsThe report states that stolen credentials and API access were used in the SwissBorg incident to initiate unauthorized transactions and drain user accounts.T1550.001 · Application Access TokenThe report states that compromised OAuth tokens in the Salesloft / Drift incident were used to access connected SaaS APIs and downstream customer data.
Products
Industries
AI platforms and tooling| **AI platforms and tooling** | 9 (15%) |Automotive| **Automotive** | 4 (7%) |Cloud services| **Cloud services** | 4 (7%) |Cybersecurity vendors| **Cybersecurity vendors** | 8 (13%) |Enterprise software| **Enterprise Software** | 9 (15%) |SaaS providers| **SaaS providers** | 5 (8%) |