AI Platforms – Cybersecurity for AI infrastructure and systems

There are four main areas of focus for securing AI models across business organisations:

Visibility

Identity and Access Management

Runtime Security

Backup and Disaster Recovery

Visibility

This is a familiar problem for all who assist customers to secure organisations in the world of cloud and SaaS. As with these technologies, AI and in particular generative AI chatbots, such as ChatGPT, are easily accessible and are therefore a significant risk in the form of shadow AI. The risk of shadow AI is vast, from loss of sensitive data via an employee uploading an internal report, to access being inadvertently granted to internal systems. This risk is highlighted by IBM; according to their research 68% of organisations do not have policies in place to prevent shadow AI. Furthermore, 21% report regulatory fines, 49% data loss, and 42% operational disruption due to shadow AI security incidents, with these incidents more than doubling since last year ⁽¹⁾.

For a holistic approach to AI security any organisation needs to start with a clear view of all AI deployed or being used.

Identity and Access Management

Identity and access management (IAM) has been a key topic for security teams and CISOs for several years, with the principle of Zero Trust Network Access (ZTNA) as a common topic in most conversations. While the principle is not new, its application to the AI world has been slow. 92% of organisation report a lack of thorough access controls for AI ⁽¹⁾. In the new AI world, organisations must adopt a zero trust, least privilege and just in time access approach to ensure governance over AI agents.

Runtime Security

Once an organisation has visibility of all AI tools in operation and has a robust identity and access framework, the next step is runtime security. With agents operating at machine speed, controls such as IAM must be continuous, with runtime visibility, verification and reporting pivotal in ensuring a robust security posture. By adopting a continuous, proactive model, organisations can be confident that AI deploys at scale, resulting in a competitive advantage while maintaining compliance with industry regulations.

Backup and Disaster Recovery

As we have seen, AI has lowered the barrier to entry for threat actors resulting in a significant increase in exploitation attempts (up 25% year-over-year ⁽²⁾) and a decrease in the time it takes for a vulnerability to be exploited. Even with complete organisational visibility, robust access and identity controls and a continuous, proactive security approach, organisations still need to be prepared for a cyber-attack. As organisations adopt AI to create efficiencies and improve competitive advantage, the key to maximising potential lies in the quality of your data. Without a robust backup and recovery protocol, a single breach could unravel all of the advantages created through AI. This is why any security strategy must include strong backup and disaster recovery programmes to contain risk.

While these guidelines provide a good overview of securing AI Platforms, this article is not exhaustive. For further insights and more complete advice please speak to your local Infinigate representative.

Dan Walker

Group Business Development Director

1. IBM’s Cost of a Data Breach Report 2026 (Cost of a Data Breach Report 2026 | IBM)
2. Fortinet’s 2026 Global Threat Landscape Report (Fortinet 2026 Global Threat Landscape Report)