v01d.bmv

Modern autonomous AI agents possess dynamic planning loops that automatically replan when an action fails. When an agent is denied access to a sensitive resource (such as payroll records), its optimization loop immediately searches for alternative tools or unmasked database views to fulfill the user's objective. Traditional stateless security gateways inspect each tool call in isolation, completely missing this tool-switching bypass. Intentfence solves this by replacing stateless inspection with an objective-level, denial-aware workflow guard. When an action under a business objective is denied, Intentfence permanently records a denial tombstone. Any subsequent action attempting to pursue that denied objective—whether via another tool, alternate resource alias, or delegated child agent—is automatically blocked at the pre-execution boundary. Unrelated legitimate work continues without false positives. Using local Meta Llama 3.2 via Ollama, IntentFence delivers zero-cost, privacy-first security advisories for SecOps teams.