Organizations and regulators are developing new security frameworks to manage the unpredictable risks posed by agentic artificial intelligence.
These measures are becoming critical because autonomous AI agents can act in ways their creators never anticipated. This unpredictability creates significant security, operational, and compliance risks for enterprises deploying the technology [1, 5].
Industry experts said that agentic artificial intelligence is creating enough risks for organizations and demands a security reframe, according to a report by Dark Reading [1]. The shift toward "agentic" AI—where systems can take independent action to achieve goals—differs from standard LLMs that simply generate text. Because these agents can interact with other software and data, they may trigger unintended chain reactions in a corporate environment [1, 2].
To address these vulnerabilities, the Monetary Authority of Singapore (MAS) has released the SAFR framework [4]. This initiative is specifically designed to tame autonomous AI agents within the financial sector, ensuring that high-stakes financial operations remain stable and compliant [4].
Private sector efforts are also emerging to bring stability to these systems. Graphwise, an AI platform provider, has established a quarterly initiative to help tame probabilistic LLMs using deterministic methods [3]. This approach aims to reduce the randomness of AI outputs, making the behavior of agents more predictable for business users [3].
Beyond specific frameworks, security professionals said they are calling for "intelligent observability" to monitor AI agents in real time [5]. This involves tracking the decision-making process of the agent to identify and stop risky behavior before it causes systemic failure [5]. The goal is to move from a model of blind trust to one of continuous verification.
“Agentic artificial intelligence is creating enough risks for organizations and demands a security reframe.”
The transition from generative AI to agentic AI marks a shift from AI as a tool to AI as an actor. While generative AI risks are primarily centered on misinformation, agentic AI introduces operational risks where a system could autonomously execute incorrect financial trades or delete critical data. The emergence of frameworks like Singapore's SAFR suggests that regulatory bodies view autonomous AI as a systemic risk that requires deterministic guardrails rather than just probabilistic guidelines.
