Operant AI Runtime Defense Platform

Operant AI, Inc. is a San Francisco company that describes itself as a runtime AI application defence platform. It was co-founded by CEO Vrajesh Bhavsar, who built iOS and macOS security technologies at Apple and the ML/AI business unit at Arm, CTO Priyanka Tembey, formerly of VMware, and COO Ashley Roof. It protects live cloud and AI applications against data exfiltration, prompt injection, AI supply chain attacks and model theft.

The platform combines several products. Endpoint Protector discovers shadow AI tools, coding agents and MCP clients on employee devices and blocks prompt injection, data exfiltration and malicious shell execution on the device. Agent Protector adds runtime monitoring, agent identity and access control, and the Semantic Firewall enforces allow, block or redact decisions based on the intent of each agent action. The Cloud MCP Gateway provides an MCP and skills registry with tool-call inspection, and AI Gatekeeper is an LLM gateway with multi-provider routing and inline redaction. It deploys to Kubernetes with a single Helm step.

• Endpoint Protector: Discovers shadow AI, coding agents and MCP clients and blocks injection, exfiltration and malicious shell commands on devices.
• Agent Protector: Monitors agents at runtime and enforces agent identity, access control and scope guards.
• Semantic Firewall: Decides allow, block or redact inline based on the intent behind each agent action.
• Cloud MCP Gateway: Provides an MCP and skills registry with tool-call inspection, blocking and full AuthNZ.
• AI Gatekeeper: An LLM gateway with multi-provider routing, request firewall and PII/PCI/PHI redaction.
• Woodpecker red teaming: Simulates real attacks safely against AI applications and agents.

A Vietnamese e-wallet can place AI Gatekeeper in front of its LLM providers to redact card and identity data and route requests by team. A bank can deploy the platform on premises to monitor agents that process payments and stop actions outside their scope. A software company can use Endpoint Protector to find unapproved coding agents and MCP servers on developer laptops.

Security teams get one platform covering employee endpoints, production agents, MCP servers and AI applications instead of a patchwork of browser extensions and legacy DLP. Inline blocking and auto-redaction protect PII, PCI, PHI, secrets and about 100 data types before they leave. Enterprise plans offer VPC, on-premises and air-gapped deployment for regulated sectors. Native agent telemetry stays under the customer's control, and the company is SOC 2 Type II compliant.

Agents reason probabilistically and act at machine speed, so pattern-based DLP and browser extensions cannot keep up; intent-based, inline controls are emerging as the answer. MCP gateways and agent gateways are becoming a distinct security category tracked by analysts. Fintechs and banks in Vietnam and Southeast Asia that let agents move money or touch card data need runtime controls that can run inside their own VPC or data centre.

Operant AI, Inc. is based in San Francisco and is SOC 2 Type II compliant. It is a contributing member of CNCF, the OWASP Foundation and the Coalition for Secure AI, is available on AWS Marketplace since October 2025, and announced a partnership with LangChain. Its researchers disclosed the Shadow Escape zero-click attack via MCP in October 2025, and its website says it is featured in five Gartner AI security reports.