Back-to-Back API Security Reports Reveal That 92 Percent of Organizations Cannot Defend Their AI Agents as Authenticated Attacks Dominate the Threat Landscape
Salt Security and KushoAI release dueling reports on the same day showing API security has become the critical blind spot of the agentic AI era, with nearly all attacks now originating from authenticated sources.
Overview
Two major API security reports released on April 8, 2026 paint a stark picture of enterprise readiness for the agentic AI era. Salt Security’s 1H 2026 State of AI and API Security report, based on a survey of 327 security professionals, found that 92 percent of organizations lack the advanced security maturity needed to defend environments where autonomous AI agents operate. A separate study from KushoAI, analyzing 1.4 million real API test executions across 2,616 organizations, concluded that one in three API failures has a direct security implication — and the most common failure category is broken authentication and authorization.
Taken together, the reports suggest that enterprises are deploying AI agents at scale while the APIs those agents depend on remain fundamentally insecure.
What We Know
The Visibility Crisis
The Salt Security report identifies a pervasive blind spot in enterprise monitoring. Nearly half of organizations surveyed (48.9 percent) reported being completely unable to observe machine-to-machine traffic, according to Salt Security. A similar proportion (48.3 percent) said they cannot effectively distinguish legitimate AI agents from malicious bots operating on their networks.
Only 24 percent of organizations maintain a fully automated API inventory, with the majority relying on partial or manual tracking methods, according to the Salt Security press release. This aligns with broader industry data: roughly 70 percent of enterprises have documentation for only 30 percent of their APIs, according to CSO Online.
Authenticated Attacks Dominate
Perhaps the most alarming finding is the shift in attack origin. Salt Labs’ threat analysis found that 99 percent of observed attack attempts now originate from authenticated sources, with 65 percent exploiting security misconfiguration (OWASP API8), according to the Salt Security press release. Rather than breaking in from outside, threat actors are exploiting legitimate credentials — increasingly those belonging to rogue agents operating with valid tokens.
The KushoAI data tells a complementary story from the testing side. Of the 1.4 million test executions analyzed, 38 percent of all security failures involved authentication and authorization issues, according to CXOToday. The study found that 91 percent of enterprise test suites verify that authentication is required on their APIs, but only 29 percent verify that access is correctly enforced across users and permissions. An API that rejects unauthenticated requests but fails to enforce cross-user boundaries is, from an attacker’s perspective, functionally open.
New Endpoints, Higher Risk
KushoAI’s analysis revealed that newly deployed API endpoints carry a 3.1 times higher authentication failure rate than endpoints older than 90 days, according to CXOToday. This suggests that the rapid pace of API deployment — two-thirds of Salt Security’s respondents reported API growth exceeding 50 percent in the past year — is outstripping security teams’ ability to properly harden new interfaces before they reach production.
The problem extends beyond first-party APIs. Only 24 percent of organizations validate third-party API responses before passing data downstream, according to CXOToday, a gap that grows more consequential as agentic systems chain together multiple API calls across organizational boundaries.
Boardroom Pressure Meets Tool Inadequacy
The disconnect between executive concern and operational capability is pronounced. Salt Security found that 78.6 percent of security leaders report increased boardroom scrutiny of AI risks, with 68.8 percent of boards specifically worried about sensitive data leakage through AI prompts or models, according to Salt Security. Yet only 23.5 percent of respondents said their existing security tools are “very effective” at preventing attacks. Nearly half (47 percent) have delayed production releases specifically because of concerns about securing APIs exposed to autonomous systems.
“Risk in the agentic era doesn’t sit in one place — it lives in how all pieces interact in real time,” Salt Security co-founder and CEO Roey Eliyahu said in the press release. “APIs are just one pillar in a much larger, deeply connected system. Most orgs aren’t ready.”
What We Don’t Know
Neither report fully addresses how the threat landscape changes when agentic AI systems begin calling APIs autonomously at scale, a scenario that remains in its early stages. Salt Security’s survey captures sentiment and posture but not real-time breach data, while KushoAI’s testing data reflects pre-production conditions that may differ from live attack scenarios.
It also remains unclear whether the industry’s existing API security standards — the OWASP API Security Top 10, last updated in 2023 — adequately capture the risks introduced by autonomous agents that chain multiple API calls, make decisions without human oversight, and operate with credentials that blur the line between machine identity and human authorization.
The question of who bears responsibility when an AI agent uses a valid API token to perform an action that no human authorized has yet to be resolved by either regulators or industry standards bodies.
Analysis
The simultaneous release of these reports underscores a structural tension in enterprise technology: organizations are deploying AI agents at a pace that their API security infrastructure was never designed to accommodate. The traditional security model — authenticate at the perimeter, then trust — fails when the entities inside the perimeter are autonomous systems capable of making thousands of API calls per minute with minimal human oversight.
The KushoAI finding that AI-generated test suites cover 2.7 times more OWASP categories than manually authored ones, as reported by CXOToday, suggests that the same AI capabilities driving the problem may also be part of the solution — if organizations invest in automated security testing rather than relying on manual processes that cannot keep pace with API proliferation.
For CISOs navigating this landscape, the data points to a clear priority: moving from authentication verification to authorization enforcement. As Sean Murphy, CISO at BECU, told CSO Online, “The API is the new perimeter” — and attackers will find unmanaged APIs before security teams do.