AI's Blind Spot Is The Link Between Data Security And Reliable Outputs

Forbes

06-05-2025

Artificial Intelligence AdobeStock_265251947
Artificial intelligence has brought tremendous change across a range of industries, super-charging productivity, speed and decision-making. However, the advances and success of AI innovation depend squarely on data security. Without secure and trusted data, AI can be susceptible to breaches, downtime, system outages, bias, misuse, and loss of public trust.
Large datasets which AI models are trained on often contain personal health, finance, or identity data. This could have serious consequences if accessed by malicious actors. The models and training data itself are also valuable intellectual property that could be a major loss if hacked.
Meanwhile, cybersecurity threats are growing as the result of powerful AI-based attacks, geopolitical attacks, ransomware and other attacks. Some 87% of security professionals said that their organization has faced AI-based cyberattacks in the last year. Of course, AI is a double-edged sword and can also be used for good to identify and defend against said attacks.
Cyber-attacks and data loss can be devastating in terms of financial and reputational damage, but downtime and system outages can also compromise the security of AI systems and impact their performance. Some 74% of technology professionals said a loss of data would be catastrophic for their business, according to a recent report. Challenges such as model inversion, model extraction and insecure data storage and transfer have led to regulation of data privacy and security requiring specific protections to maintain compliance.
'Building trust unlocks the full potential of agentic AI modules,' said Octavian Tanase, chief product officer at Hitachi Vantara. 'With a proactive approach to cyber resiliency, human-centered policies and new AI security tools, organizations can ensure that their data is a source of AI innovation, trust and growth.'
For AI applications such as agentic AI modules, data security ensures that their outputs are high quality. Without strong protections such as zero-trust architecture, intrusion detection and immutable storage, attackers can use malicious data to poison data or exploit models during inference. However, only 38% of respondents are enhancing their training data quality to explain their AI outputs, according to a recent layered report. And 24% do not even review the datasets they use to train AI models for quality, while 37% do not tag their data.
Layered Security Hitachi Vantara
Innovative agentic AI modules require not only product innovation but building trust from users, partners, and stakeholders. Without this trust, stakeholders will be less likely to participate in data sharing and collaboration which is essential for model training and ecosystem growth. When any data loss, compromise or breach occurs, it affects public confidence and can cause reputational harm to specific applications as well as the broader perception of AI.
'Traditional reactive approaches to cybersecurity are not sufficient in this new environment,' Tanase said. 'Instead, technology professionals should be proactive for data protection: a zero-trust architecture mindset and a layered approach to address technical and strategic considerations.'
Technical safeguards can provide a foundation for comprehensive data security. First, replication through synchronized copies of data in dispersed locations give protection so that critical applications stay operational and available. As part of replication, backup, and recovery act as a critical safety net, so that systems can be quickly restored in the event of failure, cyberattack, data loss or corruption.
To guard against data breaches, access controls provide user roles and permissions with granular controls to restrict data access, which reduces the risk of unauthorized entry. To enforce this access, network security such as firewalls, zero-trust architecture and intrusion detection systems block unauthorized traffic and access.
Even if data is accessed, immutable storage guarantees that data cannot be changed for a period of time. This produces a tamper-proof copy for recovery that protects against attackers altering critical data as well as any unexpected system outages. Furthermore, data encryption adds another layer of fortification by rendering data unreadable by unauthorized individuals even if they steal data. RWTH Aachen University, which teaches over 47,000 students, wanted to standardize reliable, immutable backup storage for 29 universities in its system. The University chose Hitachi Vantara for its scale and features across six locations and over 72 storage nodes.
In addition to technology tools, securing AI data today requires a focus on the central role of humans. Some 39% of business owners believe that AI needs human oversight, while 34% say AI needs more disclosure and transparency on the data it uses, according to a recent Prosper Insights & Analytics survey. To achieve this, leaders must instill a culture of awareness and accountability to drive collaboration between IT, security teams and business units. Encouraging continuous learning, improvement and knowledge sharing is essential. Data governance is critical in establishing clear policies and procedures for data classification, access control and usage. Accountability is essential so that data security is maintained for teams, tools, and vendors across the AI lifecycle of data sourcing, preprocessing, model training and deployment.
Prosper - Concern About Recent Developments in Artificial Intelligence Prosper Insights & Analytics
Fortunately, AI itself can play a critical role in data protection, providing significant new advances for defense. Automated threat detection can scan and monitor vast data stores to identify anomalies and suspicious activities in real-time.
'AI can make threat detection faster while also taking automated pre-defined actions to limit the damage, from isolating devices to halting traffic,' Tanase said. 'With these automated systems, executives can have the peace of mind that their systems can take immediate actions, while also notifying them of any potential incidents.'
AI can also be used to anonymize sensitive data while ensuring compliance with data privacy regulations. Finally, predictive security analytics use algorithms to learn from past security incidents and historical data, detect patterns and anomalies — then anticipate and help prevent threats before they occur.
Comprehensive data security is vital to the development of AI innovation. Leaders can set up their organizations for success by proactively investing in technology including AI-based options to improve protection. At the same time, building an internal culture of ownership and awareness based on data governance is a foundation. With these steps, organizations can build trust and safeguard AI innovation over the long term.