Beyond the Checklist: The Technical Backbone Taming Agentic AI

SAN FRANCISCO & ZÜRICH – July 15, 2026 – A new class of artificial intelligence is quietly moving from research labs into the core of our global infrastructure. These are not the chatbots or image generators we’ve grown accustomed to, but “agentic” AIs—autonomous systems capable of reasoning, using tools, and taking independent action to achieve complex goals. While their potential to drive efficiency is immense, they also introduce a volatile and continuous form of risk that has left most organizations dangerously unprepared. This is the complex reality that Swiss deep-tech firm LatticeFlow AI is addressing with the launch of a unified platform designed to impose order on this new, agentic world.

The company’s announcement marks a critical inflection point in our relationship with AI. It signals a move away from the static, paper-driven compliance that has characterized AI governance to date, and toward a dynamic system of continuous, evidence-based technical control. For enterprises navigating the promise and peril of autonomous systems, this shift is not just an upgrade; it’s a fundamental change in the operating philosophy for managing intelligent technology.

The Governance Gap in an Agentic World

Traditional AI governance was built for a different era. It treated risk management as a pre-deployment checkpoint: a series of assessments, documentation reviews, and ethical sign-offs before a model was released into a controlled environment. This approach is fundamentally incompatible with agentic AI. When an AI can autonomously string together tasks, access new data, and interact with other systems, risk is no longer a static snapshot. It becomes a continuous, emergent property of the system’s runtime behavior.

This creates a significant governance gap. How do you ensure accountability when an error results from a chain of decisions made by multiple autonomous agents? How do you prevent an AI from subtly drifting from its intended goals or violating privacy regulations as it learns and adapts? A recent study commissioned by SAP revealed that while investment is high, only 3% of enterprises feel fully prepared for agentic AI deployment, citing massive gaps in governance and control.

This challenge is not lost on regulatory and standards bodies. “AI governance can no longer be treated as a static verification problem,” said Dr. Apostol Vassilev, a leading expert at the National Institute of Standards and Technology (NIST). “Because we cannot build a flawless, permanent wall around AI systems, security and governance must evolve beyond cyclical, paper-driven reviews and move directly into the operational runtime to continuously measure, constrain, and manage risk.” This sentiment is the driving force behind the market’s evolution. “Only by replacing static policy reviews with continuous, runtime technical evidence can we confidently navigate the realities of the agentic world,” Dr. Vassilev added.

From Policy to Proof: A New Technical Foundation

LatticeFlow AI aims to provide that technical evidence. The company, a spin-off from the prestigious ETH Zurich, has launched what it calls a single platform to unify AI discovery, evaluation, and governance. The core innovation is its ability to connect abstract governance frameworks directly to concrete technical controls. Instead of a compliance team manually checking if a system aligns with a policy document, the platform continuously runs evaluations to generate verifiable proof.

“AI governance has long lacked a technical foundation,” explained Dr. Petar Tsankov, CEO and Co-founder of LatticeFlow AI. “There has been a persistent gap between what governance frameworks require and what organizations can actually measure. By connecting frameworks directly to technical controls, we enable enterprises to understand, control and govern AI risk with evidence, continuously.”

The platform is purpose-built for the unique challenges of agentic systems. It goes beyond generic benchmark tests by combining three key functions: use-case-specific evaluations tailored to an AI’s specific function, adaptive red teaming that simulates attacks to find vulnerabilities in agentic workflows, and continuous monitoring that re-evaluates systems as models, data, and threats change. The result is a living system of oversight that provides board-level visibility into AI risk as it emerges, not after an incident occurs.

AI Atlas: A Rosetta Stone for Global AI Compliance

Perhaps the most significant component of the new offering is AI Atlas, the world's first public registry mapping AI governance frameworks to technical risk controls. This tool acts as a Rosetta Stone, translating the dense legal and ethical language of over 40 global frameworks—including the EU AI Act, NIST AI RMF, and ISO 42001—into executable evaluations that can be run through the platform.

For any organization struggling to interpret and implement these sprawling regulations, the utility is immediate. An enterprise can select the EU AI Act within AI Atlas, and the system will present a series of ready-to-run evaluations designed to test for compliance with its specific articles. This transforms governance from a theoretical exercise into a practical, engineering discipline. It standardizes how risk is measured and provides a common language for developers, compliance officers, and executives.

This initiative represents a broader philosophical shift toward transparency and standardization in AI safety. By making the registry public, LatticeFlow AI is providing a foundational resource that can help align the entire industry on how to build and deploy trustworthy AI, moving the conversation from what should be done to how it can be proven.

Real-World Adoption and the Drive for Trust

The market appears ready for this transition. The platform is already being used by global enterprises such as SAP and energy giant Axpo, as well as fast-growing innovators in highly regulated fields. Its adoption underscores the urgent need for robust governance in sectors where failure is not an option.

“As AI moves into core investment and advisory workflows, we need continuous technical evidence that our systems are reliable, transparent and safe,” said Dr. Sina Wulfmeyer, Chief Data Officer at fintech innovator Unique AI. “The LatticeFlow AI Platform gives us that evidence, so we can deploy AI with confidence in a highly regulated environment.”

This sentiment is echoed in the banking sector. “As AI systems gain the ability to reason, use tools and take autonomous actions, policies and periodic reviews are no longer enough,” noted Dr. Holger Harms, Head of Banking Innovation Lab at Swisscom. He emphasized that technical risk controls are especially critical in banking, where innovation must scale alongside trust and regulatory accountability, calling LatticeFlow AI's work an important contribution. The industry’s need for such tools has been validated by analysts, with LatticeFlow AI being recognized in the inaugural 2026 Gartner® Magic Quadrant™ for AI Governance Platforms, a report that signals the market’s maturation.

As agentic systems become the new enterprise standard, organizations are discovering that policies and documentation alone are insufficient armor. The new imperative is continuous technical evidence—a verifiable, always-on system of checks and balances that allows innovation to scale with confidence. This is the foundation the LatticeFlow AI Platform is built to provide.