RSAC 2025 Innovation Sandbox | Knostic: Reshaping the Access Control Paradigm for Enterprise AI Security

Introduction

As generative artificial intelligence (GenAI) and large language models (LLM) rapidly penetrate corporate operations, data leakage and privacy risks have become major challenges faced by enterprises. Knostic, a startup founded in 2023, is providing enterprises with a layer of intelligent security protection with its innovative Need-to-Know access control technology to ensure the safe deployment and compliance of AI tools. Its LLM permission management system based on the “on-demand access” principle supports fine-grained data desensitization. And the “knowledge graph + federated learning” architecture, which can increase model training efficiency by 40%.

Founder Profile

Knostic was co-founded by two senior cybersecurity experts, Gadi Evron and Sounil Yu. Gadi is a recognized cybersecurity expert, serial entrepreneur, frequent contributor to industry publications and speaker at conferences ranging from Black Hat to Davos. Previously, he founded the ACoD Cybersecurity Conference, served as CEO of Cymmetria (which has been acquired), led the PwC Cybersecurity Center, and served as CISO of Israel’s National Digital Authority. Sounil is the creator of the Cyber Defense Matrix and DIE triples. Previously, he served as Chief Security Scientist at Bank of America and CISO at JupiterOne. He is a member of the FAIR Institute board and a fellow at George Mason University’s National Security Institute. The two have accumulated deep practical experience in the field of enterprise security. The company is headquartered in Reston, Virginia and Tel Aviv, Israel, and can leverage the advantages of both the Silicon Valley innovation ecosystem and the Israeli cybersecurity industry.

With deep backgrounds in finance, government and technology, the two founders are committed to solving security and compliance challenges faced by enterprises when deploying AI tools. With the increasing application of generative AI, traditional RBAC and IAM methods can no longer effectively deal with the complex data leakage risks of LLM. Especially as AI tools continue to develop, existing security protection measures are stretched and cannot effectively deal with the risks of data sharing and sensitive information leakage. Knostic has proposed an innovative “Need-to-Know” access control framework. On the basis of traditional access control, it emphasizes accurately controlling information access rights based on users’ actual work needs and business context, thereby achieving effective protection of sensitive data. This intelligent context-based access control strategy enables enterprises to minimize potential data leakage risks with the widespread application of AI tools。

Figure 1 Gadi Evron (left) founder and CEO, Sounil Yu (right) founder and CTO

Product

The birth of Knostic directly touches the core pain point of enterprise AI applications. With the rapid popularization of AI-driven enterprise search tools such as Microsoft 365 Copilot and Glean, organizations are enjoying productivity improvements while facing new risks of excessive exposure to sensitive data. Traditional identity access management (IAM) and data loss prevention (DLP) solutions have difficulty dealing with the “information inference” risk unique to LLM-even if employees do not have direct access rights, AI tools may still “infer” sensitive information that should be restricted by analyzing what they have access to. Knostic keenly captured this security gap and became the world’s first vendor to design a “need-to-know access control” framework specifically for LLM, filling a critical gap in the AI security ecosystem.

Knostic is at the intersection of AI security governance and enterprise data access control. Based on its financing history, we can see its market recognition-it completed $3.3 million in pre-seed financing in April 2024 and received another $11 million in investment in March 2025, totaling $14 million. Investors include well-known venture capital firms such as Shield Capital, Pitango First, Bright Pixel Capital, and SVCI (Silicon Valley CISO Investments), which is composed of corporate CISOs. Of particular note is the public endorsement by former NSA Director Mike Rogers as a member of Knostic’s advisory board, hinting at its potential applications in defense and security-sensitive industries.

Core Technology

The core of Knostic’s technological innovation lies in subverting the traditional binary access control model (allow/deny) and replacing it with a “Need-to-Know” decision framework based on multidimensional situational awareness. This architecture consists of three key levels:

1. The knowledge graph layer forms the cognitive basis of the Knostic system

Different from the static permission table that traditional IAM relies on, Knostic builds a dynamic enterprise knowledge graph and analyzes the semantic association and sensitivity level of data assets through natural language processing (NLP) technology. The map continuously learns data flow patterns within the organization and automatically identifies high-risk information categories such as M&A due diligence results, employee compensation structures, and unpublished product roadmaps. This ability to understand enables the system to predict which combinations of information may lead to sensitive inferences, thereby implementing preventive controls.

2. The policy engine layer implements refined access logic

Knostic introduced the concept of “Need-to-Know Policy”, which takes into account multidimensional factors such as the requester’s organizational role, task context, time sensitivity and data lineage. For example, when a marketing intern queries “quarterly sales revenue”, the system does not simply reject it, but generates an alternative response: “While the exact number is confidential, the promotions you are responsible for have contributed significantly to growth this quarter.” This response reshaping (Response Shaping) technology not only complies with the principle of least privilege, but also avoids business friction common to traditional security measures.

3. Adaptive learning layer ensures continuous evolution of the system

Knostic uses machine learning to analyze user query patterns and organizational data access trends, dynamically adjusting policy weights. When abnormal access attempts are detected (such as a sudden large number of cross-departmental data queries), the system can increase the protection level in real time and generate detailed event analysis reports for review by the security team. This adaptive capability is particularly important for preventing internal threats and permission abuse, and solves the rigidity problem of the traditional RBAC (role-based access control) model in a dynamic business environment.

From the perspective of technical implementation, Knostic adopts a lightweight API architecture that can seamlessly integrate with mainstream enterprise AI platforms such as Microsoft 365 Copilot, Glean, and Slack AI. Its deployment model supports cloud and on-premises solutions to meet the differentiated needs of different industries for data sovereignty and compliance. It is worth noting that Knostic does not simply add a filter layer around the LLM, but achieves precise blocking of information leakage paths by deeply analyzing the internal attention mechanism of the AI model. This deep integration method distinguishes it from general API gateway security products.

Solutions

Knostic has built a product matrix covering the entire life cycle of AI applications, from pre-deployment assessment to real-time protection to continuous optimization, forming a complete security closed loop. Its core product lines include:

1. Copilot Readiness Assessment

Copilot Readiness Assessment is Knostic’s flagship service, helping organizations systematically identify potential data exposure risks before deploying AI tools such as Microsoft 365 Copilot. The service scans the company’s SharePoint, OneDrive, Exchange and other data warehouses, combines automated penetration testing, generates a detailed risk heat map, and marks sensitive content areas that may be accidentally exposed by AI tools. For example, a financial institution found through an assessment that the hidden metadata in its M&A analysis documents may be obtained by unauthorized personnel through Copilot’s contextual reasoning capabilities, and then adjusted the file storage architecture and access strategy.

Figure 2 Copilot Readiness Assessment Assessment) service

2. Need-to-Know access control engine

It should be noted that the Need-to-Know Access Control Engine is Knostic’s core product. The engine monitors all queries flowing to the LLM and the responses returned in real time, executing context-aware response rewriting strategies. It is unique in that it supports multi-level response adjustment: for slightly sensitive content, only obfuscation may be performed (such as replacing specific numbers with range values); for highly confidential information, complete blocking and explanation of rejection reasons; in appropriate scenarios, “safe alternative answers” will be provided to balance business needs and security requirements. According to the case, after a retail company deployed the engine, the exposure of sensitive data through Copilot decreased by 83%, while employee satisfaction increased because the answers obtained were more business-relevant.

3. AI Entitlement Monitoring & Remediation System

AI Entitlement Monitoring & Remediation System provides continuous security. The system analyzes the usage logs of AI tools to detect abnormal patterns such as Entitlement Crawling where users use carefully designed prompt words to test the boundaries of system permissions. When high-risk behavior is discovered, the system can automatically trigger a variety of responses: from sending warning emails to temporarily freezing accounts, while generating detailed forensic reports. In addition, the system also provides a permission usage analysis dashboard to help security teams identify idle or overly broad access rights and support dynamic permission optimization based on actual needs.

Figure 3 AI Entitlement Monitoring & Remediation System (AI Entitlement Monitoring & Remediation) Function

4. Secure AI Deployment Framework

Secure AI Deployment Framework is Knostic’s solution package for industry characteristics. For highly regulated industries such as finance, healthcare, and government, Knostic provides pre-configured policy templates and compliance mapping tools to help companies meet industry regulations such as GDPR and HIPAA. For example, medical institutions can use this framework to ensure that Copilot does not leak protected health information (PHI) while still allowing doctors to efficiently access diagnosis and treatment guidelines. Knostic also provides customized training services to help customer security teams master AI-specific risk patterns and management skills, filling the gaps in traditional security knowledge systems.

Market Strategy

In the fast-growing AI security market, Knostic has established its competitive advantage through the following differentiation strategies:

1. Technical differentiation

Knostic thinks outside the box of traditional security products. Compared with static DLP solutions, Knostic’s situational awareness capabilities can more accurately identify inferred risks unique to LLM; compared with the emerging “Prompt Firewall”, Knostic not only filters malicious input, but also reconstructs output content to provide positive security value delivery. This technological positioning has enabled it to win both the RSA Conference Launch Pad and Black Hat Startup Spotlight competitions in 2024, becoming the only startup to sweep these two top security event awards.

2. Business model innovation

Knostic uses value-based pricing, which directly links its services to the revenue of its customers’ AI applications. For example, its charging model may be related to the increase in Copilot adoption or the reduction in sensitive events, rather than simply the number of users or data volume. This pricing strategy lowers the barrier to adoption for companies, especially potential customers who are evaluating the return on their AI investments. Knostic also provides flexible deployment options, from fully managed services to customer self-control solutions, adapting to the security maturity level of different organizations.

3. Ecological integration strategy

Ecological integration strategy is another key advantage of Knostic. Knostic has established deep technical cooperation with Microsoft, and its solutions have been certified by the Microsoft 365 Security Compliance Center. This collaboration ensures that Knostic can adapt to Copilot’s API changes in a timely manner and complete security assessments before new features are launched. Knostic is also actively involved in security working groups of standards organizations such as OAuth and SAML to promote industry standardization of AI security control frameworks. This ecological participation makes its solutions more interoperable and future-proof.

4. Vertical industry focus

Although its technology has cross-industry applicability, Knostic prioritizes deepening its presence in three major areas: financial services, healthcare and government. These industries not only have high data sensitivity and strict compliance requirements, but also have sufficient AI budgets. Knostic has developed a dedicated risk knowledge base for each vertical market, containing industry-specific sensitive data patterns (such as M&A code words in the financial industry, diagnostic abbreviations in the medical industry, etc.), which greatly improves detection accuracy. This verticalization strategy is also reflected in the composition of Knostic’s sales team, most of whose members have CISO or compliance officer backgrounds in target industries.

Industry Impact and Future Challenges

1. Industry Impact

The rise of Knostic reflects a profound shift in the security paradigm as AI becomes more widespread. Traditional security models are based on clear trust boundaries, while data flow in the AI era is dynamic and contextualized. The “need-to-know access” framework proposed by Knostic represents the evolution from “defensive security” to adaptive security, and its impact has gone beyond the technical level and begun to reshape the AI governance philosophy of enterprises.

In terms of technological impact, Knostic has driven the rise of AI native security control. Traditional security tools are mostly “plug-in” solutions added after the fact, while Knostic integrates security into the AI interaction process from the design stage. This concept is being emulated by more security vendors, which may give rise to a new generation of **AI security middleware** market. Knostic’s technology has also inspired deeper research into the internal mechanisms of LLM, such as how to quantify the “information inference ability” of models, which provides a new direction for academic research on AI safety.

At the organizational impact level, Knostic helps security teams transform from “business hinders” to business enablers. By providing granular access control, security departments no longer need to ban the use of AI tools across the board, but can support business departments in fully exploring the value of AI in a controlled environment. A Knostic customer case shows that after deploying Knostic, its marketing department’s Copilot usage rate increased by 40%, while security incidents decreased, showing the possibility of synergy between safety and efficiency.

2. Challenges

The technical challenges mainly come from the rapid evolution of LLM technology. New multimodal models (such as AI systems that can process images, audio and video) may introduce completely new data exposure paths, requiring Knostic to continue to expand its detection capabilities. The popularity of open source LLM (such as DeepSeek) also brings management and control difficulties. These models can be deployed locally to bypass enterprise standard security reviews. Knostic needs to develop more general detection technology rather than relying solely on the integration of specific commercial AI platforms.

Market education is another obstacle. Many companies still underestimate the data risks unique to AI, or misunderstand Knostic’s solution as a simple alternative to traditional DLP. Knostic needs to invest a lot of resources to cultivate the market and prove that its solutions are irreplaceable in the AI era. At the same time, downward economic pressure may cause companies to cut “non-core” security spending, and Knostic must more clearly quantify its ROI, such as proving value by reducing potential losses from data breaches or improving AI adoption efficiency.

Competitive pressures are increasing. Traditional cybersecurity giants (such as Palo Alto Networks, CrowdStrike) have begun to deploy in the field of AI security, and they have a more complete sales network and customer base. Knostic needs to accelerate product innovation and customer acquisition, and build sufficient market barriers before the giants fully enter the market. Its strategy of focusing on niche areas (such as Copilot security) helps to quickly establish a foothold, but it may also limit long-term growth space.

3. Outlook

In terms of technology expansion, Knostic is expected to extend its “need to know” framework from text-based LLM to multimodal AI systems, covering the increasingly rich AI application scenarios in enterprises. Developing lightweight security agents for open source models may be another direction to help enterprises manage decentralized AI deployments. Enhancing predictive security capabilities and using AI to predict potential abuse patterns will also be a focus of technological evolution.

Market expansion strategies may include deepening existing vertical industries while expanding into new areas such as manufacturing and education. Internationalization is another focus, especially in the European and Asian markets, where a high level of concern for data privacy is highly consistent with Knostic’s value proposition. Establishing more ecological partnerships similar to those with Microsoft will also accelerate market penetration.

Product evolution may see Knostic transform from an independent security product to an AI governance platform, integrating access control, compliance auditing, risk assessment and other functions into a unified console. Developing customized views for different roles (such as CISO dashboards, compliance officer reports, business administrator panels) will improve product stickiness. Knostic may also explore blockchain-based permission audit trails to enhance immutable security logs.

In short, Knostic is at the forefront of AI security changes, and its innovative “need-to-know access control” framework provides a viable path for enterprises to balance innovation and security in the AI era. Although there are many challenges, with its clear technical vision and rapid execution capabilities, Knostic is expected to grow into a key player in the field of AI security, redefining how enterprises can safely embrace the generative AI revolution.

Summary

The birth of Knostic marks that enterprise AI security has entered the era of “precise governance” from extensive control of “one size fits all”. Its dynamic access control framework based on “need to know” not only solves the real risks in LLM deployment, but also reshapes the collaborative paradigm of AI and security-security is no longer a stumbling block to innovation, but the core driving force for the large-scale application of AI in enterprises. As the wave of generative AI sweeps across the world, Knostic is expected to become an indispensable “security foundation” for enterprise digital transformation, promoting the evolution of AI technology towards a more reliable and sustainable direction.

The post RSAC 2025 Innovation Sandbox | Knostic: Reshaping the Access Control Paradigm for Enterprise AI Security appeared first on NSFOCUS, Inc., a global network and cyber security leader, protects enterprises and carriers from advanced cyber attacks..

*** This is a Security Bloggers Network syndicated blog from NSFOCUS, Inc., a global network and cyber security leader, protects enterprises and carriers from advanced cyber attacks. authored by NSFOCUS. Read the original post at: https://nsfocusglobal.com/rsac-2025-innovation-sandbox-knostic-reshaping-the-access-control-paradigm-for-enterprise-ai-security/