Fifty-three percent of respondents do not define normal behavior for machine identities, according to a report by Abnormal AI.
What Abnormal AI found about AI agent monitoring
The Abnormal AI report paints a picture of broad adoption of behavioral detection coupled with a specific blind spot for AI agents and other machine identities. More than 81% of survey respondents now rely on behavioral anomaly detection to a major or extensive degree, the report says, yet that maturity has not yet extended to AI agents or machine identities: 53% of respondents do not define what “normal” means for those non‑human actors.
Security teams report difficulty defining “normal” for agents
The difficulty is not theoretical. Abnormal AI reports that 46% of security professionals say they are "not at all effective" at establishing what normal behavior looks like for AI agents. Behavioral security systems work by learning an expected baseline for each identity, account, and system and flagging deviations; without a clear baseline for agents, the report warns, a compromised or misbehaving agent can look like any other and evade detection.
This site is the portfolio.
OSINTSights runs on Cloudflare Workers, D1, R2, and Vectorize, with an AI pipeline on Hetzner ARM. Nubivance designed, built, and operates it. We do the same for clients.
See what we buildPlanned investments tilt heavily toward agents
Despite that lack of definition, organizations are signaling where budget will flow. Abnormal AI found that 43% of respondents expect AI agents to receive their greatest behavioral‑monitoring investment over the next 24 months — more than three times the share that plans to prioritize API integrations, which stood at 14%. That planned investment suggests organizations view agents as the critical frontier for behavioral security even while many struggle to define normal agent behavior today.
Detection is trusted, but explainability and response lag
Even where anomalies are caught, operational friction remains. The report states that 90% of respondents say knowing why an anomaly triggered is critical, underscoring the importance of explainability in behavioral systems. At the same time, Abnormal AI found that 51% of organizations need four hours or more to investigate and remediate an anomaly. That gap — between the desire for clear, explainable signals and the reality of slow remediation — creates a window in which a misbehaving agent or machine identity could persist.
What this means for security teams, procurement leaders, and adversaries
- Security teams: With 46% saying they are not at all effective at defining normal behavior for agents and 53% lacking any definition for machine identities, teams will face pressure to develop baselines and tuning practices before pouring new monitoring tools into production. The report implies that explainability will be essential — 90% call it critical — so teams must prioritize signal clarity alongside coverage.
- Procurement and IT leaders: The finding that 43% expect agent-focused behavioral monitoring to receive the largest investment over the next two years signals procurement priorities. Leaders will need to balance that spend against the operational reality that many organizations cannot yet define normal agent behavior and that remediation frequently takes four hours or more.
- Adversaries and threat actors: The combination of undefined norms for machine identities (53%) and slow remediation timelines (51% requiring four hours or more) presents an exploitable window. Where baselines are absent or immature, anomalous actions by compromised agents may blend into noise, the report suggests.
The Abnormal AI report puts a fine point on a paradox: organizations are leaning into behavioral security — more than four in five rely heavily on it — even as they lack agreed baselines and rapid investigation playbooks for the non‑human identities that increasingly perform critical functions. The choices organizations make about defining “normal,” investing in agent monitoring, and speeding investigations will determine whether that investment closes a real gap or simply formalizes a new blind spot.
