Sweet Security Enhances AI Safety with New Blocking Features

On July 29, 2026, Sweet Security, a leader in proactive runtime enforcement, revealed significant advancements in AI protection with its new Agentic AI Blocking capabilities. Announced in Las Vegas at CyberNewswire, these innovations aim to enhance real-time defense against rogue AI agents that might otherwise cause harm in production environments.

Real-Time Defense for AI Enterprises

Sweet Security’s latest feature set expands its existing runtime enforcement solutions from cloud applications to AI agents. Currently, 80% of global businesses operate as AI enterprises, where autonomous agents have the potential to access sensitive data and act independently. However, the behavior of these agents often goes unchecked, leading to potential security breaches.

Many security tools today only provide alerts post-incident, leaving teams to respond after damage has already occurred. Dror Kashti, CEO and co-founder of Sweet Security, emphasized the importance of immediate decision-making to prevent unauthorized actions by AI agents. He highlighted that Sweet Security intercepts potentially harmful actions in real-time, offering a straightforward report to the team about the incident, thus enabling enterprises to adopt AI confidently.

Key Features of Agentic AI Blocking

The new capabilities of Sweet Security’s Agentic AI Blocking include terminating unauthorized tool calls and sessions during runtime, preventing the leakage of secrets, personally identifiable information (PII), and sensitive data via AI agents, and blocking live prompt injections that could misdirect an agent’s behavior.

The underlying strength of Sweet Security lies in its runtime knowledge framework, the Sweet Learning Loop. This system continuously attacks, fixes, and defends by analyzing over a billion runtime events daily, ensuring that each application and agent performs strictly within its intended purpose.

Industry Impact and Future Outlook

As the industry recognizes the necessity for runtime inspection and enforcement, Sweet Security’s proactive approach positions it as a frontrunner in AI agent security. Analysts predict that successful security products will be those that can automatically prevent risky behaviors rather than merely alerting teams post-incident.

Sweet Security’s platform is already deployed at large scales, demonstrated by its implementation at Zoomd, where it secures tens of thousands of cloud applications. Niv Sharoni, CTO at Zoomd, praised Sweet’s ability to instantly block unwanted behaviors, contrasting it with traditional tools that only alert after the damage is done.

Sweet Security’s autonomous protection system will be showcased live at Black Hat USA 2026, Booth 5721. For more information, visit their website at www.sweet.security.