Why AI security is now a patient-safety issue

A chief information security officer explains how, as AI adoption accelerates across health care, AI security is fast becoming a frontline patient-safety issue for healthcare organisations.

Healthcare organisations are rapidly embracing AI to improve care delivery, streamline operations and address workforce shortages. From clinical decision support and medical imaging analysis to patient scheduling and administrative automation, AI is increasingly embedded across modern healthcare environments.

Healthcare organisations have traditionally focused on protecting electronic health records, hospital networks and connected medical devices. AI systems introduce a new attack surface that can affect data confidentiality as well as the integrity of clinical decisions, operational processes and patient outcomes. If healthcare organisations treat AI simply as another application to secure, they risk overlooking the unique vulnerabilities these technologies introduce.

The urgency of the issue is reflected in breach data. According to the Office of the Australian Information Commissioner, the health sector accounted for 18% of all notifiable data breaches in Australia between January and June 2025, the highest of any industry.¹ As digital health systems expand and AI becomes more deeply integrated into care delivery, protecting these systems becomes even more critical.

AI expands the healthcare attack surface

Healthcare data has always been a prime target for cybercriminals. Protected health information (PHI) is highly valuable, and healthcare environments often combine legacy systems, modern cloud platforms and large user populations across clinical and administrative teams.

AI expands that already challenging attack surface in a number of ways:

1. AI systems depend on vast datasets. These datasets, which often contain sensitive patient information, are used to train and refine models. If attackers gain access to training environments or manipulate the data feeding AI systems, they may be able to compromise both privacy and accuracy.
2. Many AI systems interact with users through natural language interfaces or automated workflows. These systems can be vulnerable to techniques such as prompt injection, where attackers craft inputs designed to manipulate the model’s behaviour.
3. AI models themselves can become targets. Through techniques such as model manipulation or model inversion, adversaries may attempt to extract sensitive data or influence model outputs.

At the same time, the broader cyberthreat landscape is intensifying with more exploitation attempts, demonstrating the scale at which attackers are probing organisations for weaknesses. In health care, where digital systems increasingly support clinical decisions and operational workflows, these risks can have far-reaching consequences.

When cybersecurity becomes patient safety

Traditional cyber incidents in health care typically focus on system availability or data exposure. For example, ransomware attacks disrupt hospital operations and delay care delivery. AI introduces the potential to affect the integrity of medical insights and clinical workflows.

If an AI model used to analyse imaging data is manipulated, diagnostic results could be affected. If an AI system supporting triage or scheduling is compromised, patient prioritisation may be disrupted. Even administrative AI tools handling sensitive data could expose patient records if security controls are inadequate.

This means that the impact of AI security failures may extend beyond privacy and compliance into direct clinical risk. Cybercriminals are also becoming faster and more automated. Global reconnaissance scanning has increased, highlighting how attackers increasingly use automation to identify vulnerable systems before organisations can patch them. This makes AI security a patient safety and operational resilience issue for healthcare leaders, not just the IT department.

Compliance alone is not enough

Healthcare organisations already operate within strict regulatory frameworks governing patient privacy and data protection. However, many of these frameworks were designed around traditional IT systems rather than AI-driven decision environments. Simply extending existing security controls to AI platforms may not be sufficient.

AI systems require new governance approaches that address how models are trained, validated, monitored and secured throughout their lifecycle. Without these controls, healthcare organisations risk deploying technologies that introduce unseen vulnerabilities. The challenge is that many healthcare providers are adopting AI faster than they can build the governance frameworks needed to manage it securely.

Meanwhile, cybercriminal ecosystems continue to expand. In underground markets, compromised credentials and corporate access are increasingly traded as commodities, lowering the barrier for attackers to infiltrate enterprise networks. For healthcare organisations managing vast volumes of sensitive patient data, this growing cybercrime economy increases the risk of targeted attacks.

**************************************************

Building AI security into healthcare strategy

To safely realise AI’s benefits, healthcare organisations should take a proactive approach to AI security and governance.

1. Establish AI governance frameworks and standards

Healthcare organisations need clear policies defining how AI systems are developed, deployed and monitored. Governance frameworks should address issues such as training data management, model validation, access control and auditability. Healthcare organisations should also look to formal standards, such as the ISO 27090, which is currently in development. Security and clinical leaders should collaborate to ensure AI tools meet both cybersecurity and patient safety standards.

2. Secure the data pipeline

AI models are only as trustworthy as the data used to train and operate them. Healthcare organisations should protect training datasets with strong access controls, encryption and monitoring to prevent tampering or unauthorised access. Data integrity checks can also help detect attempts to manipulate AI training inputs.

3. Strengthen identity-centric security

Many AI risks arise from unauthorised access to systems, datasets or development environments. Implementing strong identity and access management, including multi-factor authentication and least-privilege access, helps reduce these risks. Healthcare organisations should also ensure AI platforms are integrated into broader identity security frameworks.

4. Monitor AI behaviour and outputs

Traditional security monitoring focuses on networks and endpoints. AI systems require additional oversight to detect abnormal model behaviour, unexpected outputs or attempts to manipulate interactions. Continuous monitoring helps organisations identify emerging threats and respond quickly.

5. Align cybersecurity with clinical resilience

Healthcare organisations should treat AI security as part of their broader resilience strategy. Security teams, IT leaders and clinical stakeholders must work together to ensure AI systems support, not undermine, care delivery.

**************************************************

Securing innovation in health care

Artificial intelligence holds enormous promise for health care. It can improve diagnostics, enhance operational efficiency and help clinicians focus more time on patient care. However, as AI becomes embedded in healthcare infrastructure, the consequences of security failures grow more significant.

Healthcare organisations must recognise that AI security is no longer just about protecting technology; it’s about protecting patients. By building strong governance frameworks, securing data pipelines and integrating AI into broader cybersecurity strategies, healthcare leaders can ensure innovation moves forward safely without compromising trust.

^{1. Latest Notifiable Data Breach statistics for January to June 2025. Office of the Australian Information Commissioner (OAIC). Accessed 27 March, 2026. https://www.oaic.gov.au/news/blog/latest-notifiable-data-breach-statistics-for-january-to-june-2025}

*Cornelius Mare is Chief Information Security Officer, Australia at Fortinet.

Top image credit: iStock.com/sturti