The rapid evolution of artificial intelligence has moved beyond the era of static conversational models into a phase defined by autonomous agents that can act, decide, and execute complex workflows without constant human supervision. As enterprises integrate these “doers” into their core operations, the security landscape faces an unprecedented challenge: how to govern entities that possess the authority to modify production code and access sensitive data. Manifold, a cybersecurity startup, recently addressed this gap by securing eight million dollars in seed funding to pioneer an Agentic AI Detection and Response platform. Led by Costanoa Ventures and supported by seasoned experts like the former security heads of Uber and Google DeepMind, this investment signals a shift in priority toward securing the autonomous layer. While tools like GitHub Copilot and Claude Code have become staples for developers, they also introduce significant blind spots that traditional endpoint defenses were never designed to manage.
Bridging the Gap in Autonomous Oversight
From Monitoring Text: Analyzing Actions
Traditional approaches to artificial intelligence security have focused almost exclusively on the inputs and outputs of large language models, employing firewalls and prompt-monitoring tools to catch malicious intent or data leaks. However, these first-generation methods are inherently reactive and limited because they only interpret the text an agent generates, remaining blind to the actual operations the agent performs on a local machine or network. Manifold addresses this by shifting the focus toward runtime visibility, where the primary objective is to monitor the specific actions an agent takes after receiving a command. This transition is critical because an agent might provide a benign-sounding response while simultaneously executing a script that could compromise internal security. By prioritizing action over language, the platform offers a layer of defense that treats AI agents as active system participants rather than simple text generators, ensuring every operation is logged and analyzed in real-time.
Seamless Deployment: Maximum Visibility
One of the most significant barriers to adopting new enterprise security software is the friction associated with complex architectural overhauls and the deployment of intrusive endpoint agents. Manifold overcomes this challenge by utilizing an agentless deployment model that integrates directly into existing corporate infrastructure, allowing security teams to achieve full visibility across their environment in a matter of days. This approach is designed to be as non-disruptive as possible, ensuring that the introduction of governance tools does not slow down the development cycles or the productivity of knowledge workers who rely on AI assistants. By operating at the infrastructure level rather than requiring individual installations on every machine, the platform can scale rapidly alongside a company’s growing fleet of autonomous agents. This streamlined integration allows organizations to focus on leveraging AI capabilities while the security layer operates silently in the background, providing the necessary guardrails.
A Vision for Runtime Governance
Leveraging Specialized: Security Expertise
The leadership behind this new security paradigm brings a wealth of specialized expertise to the table, having previously developed the most widely utilized open-source firewalls for large language models. Their experience with foundational AI security tools revealed a fundamental limitation: relying solely on natural language classification to judge the intent of an AI is often an exercise in futility. In practice, this approach frequently results in a flood of false positives that can overwhelm security departments and lead to alert fatigue, where critical threats are ignored amidst the noise. By recognizing that language is a poor proxy for actual system behavior, the Manifold team has pivoted toward a strategy that prioritizes the context of system interactions. This shift acknowledges that as AI models become more sophisticated, they will inevitably find ways to bypass simple keyword filters or intent-based classifiers, making it necessary to monitor the tangible outcomes of their autonomous decisions.
Securing the Next Layer: Enterprise Infrastructure
As autonomous agents gain the capability to browse the open web, modify production code, and interact with third-party services, the attack surface of the modern enterprise is expanding at an unprecedented rate. These agents are no longer just passive assistants; they are becoming the next major layer of enterprise infrastructure, comparable to the shift toward cloud computing or mobile technology in previous decades. Investors and industry experts anticipate that specialized security solutions for this agentic layer will soon move from being an optional luxury to a mandatory component of the standard corporate technology stack. The risks associated with unmanaged autonomous activity—ranging from accidental data deletion to targeted external attacks—are simply too high for most organizations to ignore. This expansion necessitates a new category of defense that is specifically tuned to the unique lifecycle and behavioral patterns of AI agents, which differ significantly from human users.
The successful introduction of this detection and response platform addressed a critical vulnerability that had emerged during the rapid integration of autonomous AI into the modern workspace. By bridging the gap between traditional endpoint security and AI governance, the framework provided a foundational layer that allowed companies to transition from using AI as a simple thought assistant to a reliable tool for automated action. Organizations that adopted these measures found themselves better equipped to manage the complexities of a hybrid workforce where human and digital agents collaborated on sensitive tasks. The path forward required security leaders to move beyond legacy monitoring and embrace runtime governance as an operational imperative for maintaining integrity. Ultimately, the shift toward behavioral mapping and real-time oversight established a new standard for trust, ensuring that the pursuit of efficiency through AI did not come at the expense of organizational security.
