Bottom line: Sweet Security blocks unauthorized agent actions in real time rather than detecting them retroactively – without disrupting production systems.
Sweet Security has announced Agentic AI Blocking – a technology that stops deviant behavior from AI agents at runtime before damage occurs. The company addresses a core risk of enterprise AI: agents that access data without authorization or are manipulated through prompt injections.
Sweet Security, founded in 2023 and funded with $120 million by Evolution Equity Partners, Munich Re Ventures, and other investors, is expanding its runtime enforcement platform for cloud and AI systems. The new Agentic AI Blocking capability prevents AI agents from performing unauthorized actions at runtime – such as terminating unauthorized tool calls, blocking exfiltration of secrets or personal data, or interrupting prompt injection attacks before they divert the agent from its intended course.
According to Sweet, 80 percent of enterprises worldwide are AI companies. These deploy agents that assume identities and independently access data – without systematically verifying whether their actual behavior matches the intended behavior. Traditional security tools detect and alert only after the fact. Sweet, by contrast, focuses on preventive blocking at the moment an agent performs an unauthorized action – effectively eliminating the window between intent and harm.
The technical foundation is the Sweet Learning Loop, a continuous analysis of over one billion daily runtime events. The system learns what behavior each application and agent is intended to perform, and enforces the principle “do exactly what your creator intended, and nothing more.” This approach enables enforcement without impacting live production systems.
Niv Sharoni, CTO at Zoomd, where Sweet protects tens of thousands of cloud applications, describes the difference: while most tools make harmful behavior visible only in retrospective reviews, Sweet blocks it at runtime as soon as it occurs. This distinguishes risk monitoring from actual risk elimination. Industry analysts now expect that winners in agent security will be products that automatically prevent deviant behavior, rather than simply generating more dashboards and alerts.
Source: www.csoonline.com · Published July 29, 2026
Lumi AI News — AI-assisted curation in accordance with Article 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.