The EU AI Act enforcement deadline of August 2026 is fast approaching, yet Italian companies remain dangerously unprepared. A joint analysis by Deloitte and the Cloud Security Alliance (CSA) covering more than 100 organizations active in Italy reveals a stark reality: 86% of Italian firms have allocated zero budget specifically for GenAI security, while regulatory penalties for non-compliance loom just eighteen months away.
This is not merely a budget problem—it reflects a fundamental maturity deficit in how Italian enterprises approach generative AI governance and risk management.
Why This Matters:
The stakes are tangible and immediate:
• 86% of Italian firms have allocated zero budget specifically for GenAI security.
• Only 9% of Chief Information Security Officers have full visibility into AI initiatives within their own companies.
• The EU AI Act enters operational force in August 2026, bringing mandatory compliance requirements and substantial penalties for non-conformance.
• Shadow AI — unauthorized employee use of generative tools — is exposing sensitive corporate data at scale, with roughly 1 in 31 prompts posing a concrete data leakage risk.
The EU AI Act: What Italian Companies Must Know by 2026
The EU AI Act, formally designated as Regulation (EU) 2024/1689, transitions from policy framework to enforceable law in August 2026. It introduces a risk-based classification system for AI applications, with mandatory requirements for high-risk systems including:
• Structured technical documentation and audit trails.
• Continuous risk assessments and human oversight mechanisms.
• Integration of AI governance into internal control systems.
Non-compliance carries substantial financial penalties that can reach up to 6% of global annual revenue for the most serious violations. The Act applies extraterritorially, meaning any provider — Italian or foreign — serving users in the EU must conform.
Parallel to this, the NIS2 Directive (Directive (EU) 2022/2555), effective since October 2024, imposes stricter cybersecurity obligations on essential service operators and important entities across sectors. While not AI-specific, NIS2 indirectly mandates robust security for AI systems embedded in critical infrastructure, supply chains, and customer-facing services.
For Italian organizations, this means compliance is not optional—it is a legal requirement with material financial consequences.
The Maturity Deficit
The Deloitte-CSA survey, conducted across Italian organizations in 2025 and released this year, reveals that two-thirds of companies lack a formalized AI security strategy. Among those that have begun planning, just 9% report a fully implemented framework, and a mere 3% possess mature governance models coupled with operational capabilities to manage AI-specific threats.
This is not a problem of awareness in isolation. Fabio Battelli, enterprise security leader at Deloitte Italy, notes that almost 90% of organizations manage AI risk outside formal frameworks, relying instead on ad-hoc measures or traditional cybersecurity architectures that were never designed to handle the unique threat landscape of generative systems.
Daniele Catteddu, chief technology officer at the Cloud Security Alliance, frames the challenge bluntly: the divergence between deployment velocity and governance capacity now ranks among the most pressing issues facing Italian security leaders.
Blind Spots and Budget Gaps
Visibility remains a critical weakness. The survey found that only 9% of CISOs claim comprehensive oversight of GenAI projects across their enterprises. This lack of transparency feeds directly into the phenomenon of shadow AI, where employees adopt external generative tools — often consumer-grade platforms — without organizational approval or security vetting.
The result: 88% of organizations using GenAI tools regularly face exposure incidents, with sensitive information inadvertently shared with third-party systems beyond corporate control.
On the financial front, the picture is equally concerning. 86% of Italian companies have yet to allocate any dedicated budget for GenAI security. In the minority of firms that have, spending averages just 2-3% of total cybersecurity expenditure. This comes even as the broader Italian cybersecurity market reached €2.78 billion in 2025, up 12% year-on-year, and 70% of large enterprises plan to increase overall security budgets in 2026.
Key barriers cited by respondents include limited technical understanding of AI-specific risks (68%), scarcity of specialized personnel (59%), and general budget constraints (45%). Compounding the skills gap: 63% of organizations have not yet launched structured training programs focused on GenAI security.
Understanding AI-Specific Security Threats
Italian firms express concern over data loss, privacy violations, and shadow AI, yet many appear less attuned to threats intrinsic to generative models themselves. Understanding these risks is essential for compliance:
• Prompt injection attacks: Malicious inputs can manipulate AI systems to execute unintended actions or bypass security controls. Think of it as tricking the AI into ignoring its original instructions.
• Model and data poisoning: Attackers corrupt training datasets or alter how the AI "thinks," introducing vulnerabilities or causing the system to produce biased or incorrect outputs.
• Algorithmic hallucinations and bias: AI systems can generate plausible-sounding but false information, or exhibit discrimination in decision-making—risks that regulatory frameworks now explicitly address.
• Excessive agent autonomy: AI systems acting beyond intended parameters can cause unintended business disruptions or expose sensitive information.
Critically, 97% of Italian organizations have not adapted DevSecOps methodologies — which integrate security testing into development workflows — to account for these GenAI-specific vulnerabilities. Instead, they rely on legacy processes unsuited to the dynamic, API-driven architecture of modern AI applications.
The threat environment is intensifying. In 2025, 34% of large Italian enterprises reported cyberattacks with significant recovery costs, and 3% experienced operational disruptions affecting production. Industry data indicates that AI-enhanced attacks in Italy surged 90% compared to the prior year. Separately, 54% of organizations globally have either confirmed or suspect AI-related security incidents, with an additional 24% lacking sufficient telemetry to verify suspicious activity.
Immediate Pathways to Compliance
Industry experts and regulatory bodies recommend several immediate actions for Italian organizations:
Establish formal AI security governance: Companies should define clear accountability structures, integrating security teams into AI project design from inception. Currently, 45% of Italian firms lack any dedicated oversight body for GenAI, and 13% have governance committees that exclude security leadership. The Agenzia per la Cybersicurezza Nazionale (ACN) offers guidance documents and resources through its official portal (www.acn.gov.it) to help organizations establish governance frameworks aligned with national and EU standards.
Increase dedicated investment: Even modest allocations — 5-10% of cybersecurity budgets — can fund critical initiatives such as AI-specific threat detection, model integrity monitoring, and secure API gateways. Given that the broader Italian cybersecurity market is expanding, investment in GenAI security is increasingly viewed as a competitive advantage rather than a cost center.
Address the skills gap: Launch training programs targeting both technical staff and business units. This includes educating employees on the risks of shadow AI and safe GenAI usage. The Agenzia per l'Italia Digitale (AgID) provides resources and training modules focused on digital governance and cybersecurity practices for public and private organizations.
Adopt recognized frameworks: The NIST AI Risk Management Framework, ISO/IEC 42001 (AI management systems), and ENISA's Framework for AI Cybersecurity Practices (FAICP) provide structured approaches to operationalizing AI Act requirements and building resilient governance models. While voluntary, these standards are increasingly viewed as compliance accelerators and are widely recognized by Italian regulators and industry bodies.
Implement AI-aware security practices in development: Integrate security testing for prompt injection, model poisoning, and adversarial inputs into development workflows. Establish continuous monitoring for model drift, data integrity, and output anomalies. This aligns with both AI Act requirements and broader NIS2 obligations.
Manage shadow AI proactively: Deploy policy frameworks and technical controls to detect unauthorized GenAI tool usage. Clear usage guidelines, combined with enterprise-approved alternatives, can reduce risk without stifling innovation. Many Italian cybersecurity vendors now offer shadow AI detection tools specifically designed for compliance with Italian data protection and privacy standards.
Italian Institutions Leading the Way
Some Italian institutions are moving ahead of the curve, signaling what proactive AI governance looks like. The Banca d'Italia has initiated consultations with AI vendors to assess security risks in the financial sector, establishing sector-specific standards that other organizations are beginning to reference. Italian cybersecurity firms like Tinexta Cyber and IT4LIA AI Factory are offering compliance services tailored specifically to GDPR, AI Act, and NIS2 requirements, including bias detection, data protection, and model security audits. Additionally, industry associations such as Confindustria and Assintel have begun publishing guidance documents and hosting webinars on AI governance for their members.
These early movers demonstrate a critical truth: as the August 2026 enforcement date approaches and cyber threats intensify, the cost of inaction is becoming untenable. With generative AI now a fixture in customer service automation, document analysis, and business intelligence across Italian enterprises, the question is no longer whether to secure these systems, but how quickly organizations can close the maturity gap before it becomes a regulatory penalty or a competitive liability.
For Italian business leaders and security professionals, the time to act is now—not in 2026.
