Botnets, adversarial attacks, and data poisoning: the AI threats companies fear the most

AI-Directed Botnets and Adversarial Attacks Among Most Concerning Threats for Companies

Companies are heavily investing in artificial intelligence but recognize they are less prepared to face attacks targeting AI systems compared to other cyber threats. A PwC survey of 3,934 business leaders in 71 countries reveals that 50% of security and technology executives rank AI system attacks among the top five preparedness gaps.

Among the most feared threats, three specific types emerge: AI-controlled botnets capable of large-scale action, adversarial attacks that manipulate inputs to obtain incorrect responses, and data poisoning, with the insertion of misleading records into training datasets. PwC's assessment highlights how advanced AI models can now identify and exploit unknown software vulnerabilities with minimal human intervention.

Quick Response

Increasing Budgets, but Insufficient Continuity Plans

84% of security and finance leaders surveyed predict an increase in cybersecurity budgets, with AI among the main priorities. However, only 39% have developed complete business continuity plans, while nearly a quarter of companies are not developing any formal plan. PwC defines cyber incidents as a matter of "when," not "if".

The Quality of Data: AI's Achilles' Heel

Data security measures prove insufficient: on average, companies have implemented only 3 of the 7 enterprise-level data risk measures. Only about half have adopted data classification systems, a fundamental step to identify sensitive information. PwC emphasizes that the reliability of AI directly depends on the quality and security of the data it is based on.

Limitations on AI Agents' Autonomy

Less than a quarter of companies would allow AI agents to autonomously contain and resolve attacks without human approval. The majority limits these agents to low-risk actions or maintains human control. Over half of the leaders interviewed cite reliability and technological maturity as the main barriers to adoption.

Matt Rowe, Chief Security Officer of Lloyds Banking Group, states: "To succeed in the AI era, we must be able to defend ourselves at machine speed".

The Challenge of AI Governance

Companies show different approaches in managing AI risk: a third have created dedicated roles such as the Chief AI Officer, while others assign this responsibility to technological functions or the CISO. PwC recommends that when the outcome must always be correct, the work of AI agents should go through independent checks such as human reviews or approval workflows.

The Importance of Cyber Insurance for Modern Companies

In this context of growing threats and complexity, a cyber insurance becomes an essential tool to mitigate financial risks arising from cyber incidents. Cyber risk policies offer coverage for direct and indirect losses, incident response costs, and potential legal liabilities.

The Proactive Approach: Managed Detection and Response

Given the threatening landscape, a MDR service (Managed Detection and Response) represents a strategic solution for companies. These services offer 24/7 monitoring, advanced threat detection, and rapid incident response, bridging the skills gap that many organizations face.

The Need for NIS2 and DORA Compliance

In this scenario, NIS2 compliance and adherence to the DORA regulation become legal and operational imperatives. These regulations strengthen requirements for digital operational resilience and protection of sensitive data, crucial aspects for companies implementing AI solutions.

The Importance of Penetration Testing and Vulnerability Assessment

In a context where AI-directed attacks are becoming increasingly sophisticated, regular penetration testing and accurate vulnerability assessment are fundamental to identify and mitigate vulnerabilities before they can be exploited. These tests provide an independent evaluation of corporate defenses and help prioritize security interventions.

The Challenge of Protecting Sensitive Data

With the increase in attacks aimed at corrupting training datasets, an enterprise DLP system (Data Loss Prevention) becomes a crucial tool for protecting sensitive data. These solutions monitor and control data access, preventing the loss of critical information and reducing the risk of data poisoning.

The Importance of a Managed SOC

Given the complexity of modern threats, a managed Security Operations Center offers companies advanced monitoring and incident response capabilities. These specialized services provide access to advanced technical skills and cutting-edge security technologies, essential to address AI-enabled threats.

The Need for Robust Identity Management

In a context where cyber attacks are becoming increasingly sophisticated, an identity management (IAM) system becomes an essential component of the security strategy. These systems ensure that only authorized users can access critical resources, reducing the risk of unauthorized access and internal attacks.