Skip to content
IT
Cybersecurity

Closed vs. Open AI Models: How to Choose the Safest Strategy for Your Company

Multi-model architectures, structured data, and security fundamentals: the keys to resilient AI implementation

Published on By Sebastiano, GoYou.it 6 min read
Listen to the full article Full reading · 10:08 · from the channel GoYou Cybersecurity
Closed vs. Open AI Models: How to Choose the Safest Strategy for Your Company

KEY TAKEAWAYS

  • The choice between closed and open AI models requires an in-depth analysis of the organization's specific needs
  • A multi-model architecture can optimize agility, costs, and resilience in AI implementation
  • The structure of data and its preparation are critical before integrating AI into existing architectures
  • Consolidating business security fundamentals is the most effective step to manage AI-related risks

The choice between closed and open AI models is not binary

The discussion about which approach to artificial intelligence is more reliable - closed systems or open-weight - does not have a definitive answer. Morgan Adamski, leader of the Cyber, Data & Technology Risk practice at PwC, emphasizes that considering a model, tool, or vendor as the definitive solution can limit organizations' adaptability.

Quick Answer

Organizations must evaluate both closed and open-weight models to build flexible architectures. The choice depends on specific factors such as agility, cost efficiency, and resilience. It is crucial to prepare the data structure before implementing AI solutions, focusing on enterprise security fundamentals.

Multi-model architectures for greater agility

Adamski highlights how a multi-model architecture can improve operational agility, cost efficiency, and overall resilience. This strategy allows organizations to quickly adapt to technological evolution, avoiding the risk of "analysis paralysis" that blocks many strategic decisions.

The critique of the "Frankenstein" architecture

Another critical point is the preparation of the data structure for AI. Many organizations make the mistake of implementing AI solutions on existing non-optimized architectures, creating an unstable hybrid system that Adamski defines as "Frankenstein". This approach can generate security vulnerabilities and operational inefficiencies.

Mastering AI security fundamentals

According to the expert, the most important step for organizations is to master enterprise security fundamentals specific to AI. This includes risk management, data governance, and the implementation of appropriate security controls before scaling AI solutions.

The importance of flexibility in model choice

Adamski emphasizes that the choice between closed and open-weight models is not a definitive decision, but part of a broader strategy. Organizations must maintain flexibility to adapt their AI solutions to changing needs, while preserving security and innovation capabilities.

Key resources for further exploration

For those who want to explore these topics further, PwC offers in-depth resources on AI-powered cybersecurity and leadership in the era of AI-exposed risk. These analyses provide concrete insights for implementing effective security strategies in complex business contexts.

The role of the AI Security Brief

The AI Security Brief podcast, where this topic was discussed, represents a reference point for security and technology leaders. Each episode addresses trends, threats, and urgent decisions in the AI landscape, offering practical insights to stay competitive.

TrendAI: proactive security for innovation

TrendAI™ offers proactive security solutions designed to eliminate risks and foster innovation. Their mission is to help organizations navigate the complex AI landscape with confidence, ensuring maximum protection of data and infrastructure.

The regulatory landscape and implications for companies

The evolution of AI architectures is set within a rapidly transforming regulatory context. The NIS2 directive, which came into force in January 2023, imposes stringent NIS2 compliance requirements for organizations operating in critical sectors. Companies implementing AI solutions must now integrate the security controls required by the regulation, with particular attention to risk management and digital operational resilience.

The impact on cyber insurance costs

The choice between closed and open-weight models directly affects the cyber insurance premium. Insurance companies are indeed differentiating premiums based on the robustness of the AI architectures employed. According to a recent PwC report, organizations with well-designed multi-model architectures can benefit from premium reductions of up to 15%, while "Frankenstein" systems increase the risk of breach remediation and thus the cost of policies.

The challenges of identity management

The adoption of multi-model AI architectures raises new identity access management issues. Each model indeed requires specific data access, creating complexity in managing permissions. According to Adamski, 60% of breaches related to AI implementations occur due to errors in access configuration. This highlights the importance of integrating Data Loss Prevention solutions from the early design stages.

The role of Managed SOC

The increasing complexity of AI architectures is pushing many organizations to consider Managed Security Operations Center solutions. MDR providers are developing specific offerings to monitor AI implementations, with particular attention to open-weight models that require continuous vigilance. PwC has estimated that 70% of large companies will adopt managed services by 2026 to address new AI-related threats.

Integration with legacy systems

An often overlooked crucial aspect is the integration between AI models and legacy systems. According to PwC, 40% of organizations adopting IA solutions make the mistake of not adapting old systems, creating security vulnerabilities. This requires a structured approach to business continuity, with particular attention to enterprise backups and disaster recovery as a service plans.

Opportunities for CISOs

For Chief Information Security Officers, the evolution of AI architectures represents an opportunity to strengthen their strategic role. The ability to design and manage multi-model architectures has indeed become a key competence to guide the organization through digital transformation. According to Adamski, CISOs who master these aspects can increase their influence at the board level, aligning AI security with business objectives.

The future of AI architectures

Looking ahead, the adoption of multi-model architectures is set to grow exponentially. According to PwC forecasts, by 2027 65% of large companies will use at least three different AI models, with a significant impact on risk management and data governance. This scenario requires a proactive approach to DORA compliance, with particular attention to operational resilience and vulnerability management.

The need for specialized skills

The evolution of AI architectures is creating a growing demand for specialized skills. According to PwC, the cybersecurity job market will see a 30% growth by 2025, with particular demand for experts in penetration testing and vulnerability assessment specific to AI solutions. This trend underscores the importance of investing in continuous training for security teams.

Implications for wealth management

The adoption of multi-model AI architectures also raises issues of crypto wealth management. Organizations using AI models for wealth management must indeed address challenges related to institutional crypto custody and crypto taxation in Italy. PwC has identified a 40% increase in requests for consulting in this area, with particular attention to crypto AML compliance.

Towards a trust ecosystem

Finally, building flexible and secure AI architectures requires the creation of a trust ecosystem. This implies collaboration between vendors, managed service providers, and user organizations. According to Adamski, only through a collaborative approach will it be possible to address the challenges related to AI security, ensuring innovation and sustainable growth at the same time.

Frequently Asked Questions

What is the average cost of an MDR service specialized in AI?

The average cost of an MDR service specialized in AI ranges between €50,000 and €200,000 annually, depending on the complexity of the architecture and the number of models managed. Costs can increase significantly for organizations operating in regulated sectors.

How do implementation costs compare between closed and open-weight models?

Closed models generally have higher licensing costs, with a range from €50,000 to €500,000 annually for large organizations. Open-weight models, although free, require significant investments in infrastructure and specialized skills, with costs that can exceed €200,000 for the first implementation.

What are the minimum NIS2 compliance requirements for AI implementations?

NIS2 compliance requirements include risk identification and management, incident notification, supply chain assessment, and the adoption of appropriate technical and organizational measures. Organizations must also demonstrate the ability to ensure digital operational resilience.

Editorial Note and Disclaimer

The guides and content published on GoYou are the result of independent research and analysis activities, for informational, educational, and in-depth purposes.

GoYou does not constitute a journalistic publication or an editorial product pursuant to Law No. 62/2001 and does not provide real-time information.

The GoYou project does not provide professional, technical, legal, or financial advice and disclaims any liability for the improper use of the information published.

In the Crypto sector, every investment involves risks: readers are invited to always inform themselves independently before making any decision.

Source: thecyberwire.com · GoYou.it is not a registered news outlet; all content is for informational and educational purposes.

Listen GoYou Cybersecurityevery article read in full
Go to the channel

You may also like

Behind every article is the work of our passion. If you enjoy GoYou.it, you can support its development and continuity.

Support with PayPal