Starting September 11, 2026, the regulatory landscape for AI developers shifts to a mandatory reporting schedule for actively exploited vulnerabilities. Under EU CRA Article 14, manufacturers of products with digital elements (PDEs) must adhere to a strict disclosure timeline for actively exploited vulnerabilities. This mandate applies to any entity placing such products on the EU market, regardless of where the company is headquartered. The clock begins the moment a manufacturer becomes aware of an active exploit, triggering a 24-hour window for an early warning submission to the ENISA Single Reporting Platform (SRP).
The requirements under Article 14 are granular and unforgiving. Beyond the initial 24-hour early warning, manufacturers must provide a comprehensive vulnerability notification within 72 hours and a final report within 14 days of implementing corrective measures. These reports are directed to ENISA and the relevant national CSIRT designated as a coordinator. While this is not a continuous monitoring duty, the inability to detect an exploit can be used as evidence of non-compliance. The definition of an actively exploited vulnerability, per Article 3(42), requires reliable evidence of unauthorized malicious exploitation, setting a specific threshold that is narrower than the standard disclosure of common vulnerabilities and exposures.
AI agents, MCP servers, and inference endpoints fall squarely within the CRA’s technology-neutral definition of products with digital elements. This classification is critical because it subjects these modern AI architectures to the same rigorous reporting standards as traditional software. The CRA does not distinguish between human-authored and AI-generated code, placing the burden of responsibility entirely on the manufacturer. As these systems become more integrated into enterprise workflows, the scope of what constitutes a reportable vulnerability expands to include agent-specific threats such as prompt injection, tool-call hijacking, and agent identity compromise.
A significant compliance ambiguity arises from the current disclosure vacuum. The existing CVE and CWE infrastructure lacks dedicated entries for prompt injection and agent-native attack patterns. Manufacturers are therefore forced to navigate a reporting regime where the technical taxonomy for their primary risks is not yet standardized. This creates a scenario where a manufacturer might be aware of an exploit but lack the formal industry-standard classification to categorize it, potentially complicating the 24-hour reporting requirement. The burden remains on the builder to identify and report these incidents despite the lack of mature industry-wide tracking mechanisms.
The extraterritorial reach of the CRA ensures that US-based AI companies shipping products to the EU are fully bound by these obligations. Furthermore, Article 69(3) introduces retroactive coverage, meaning that all in-scope products placed on the market before December 11, 2027, are subject to these reporting rules. This includes products already shipped and currently in operation. Non-EU manufacturers are required to designate an authorized representative within the EU to manage these obligations, ensuring that the regulatory reach extends effectively across borders to all participants in the European market.
This new reporting layer operates in parallel with existing frameworks, creating a complex web of overlapping obligations. The EU AI Act, which saw its high-risk obligations enter into application on August 2, 2026, functions alongside the CRA, as do the NIS2 incident reporting requirements, SEC 8-K Item 1.05, and CISA CIRCIA. There is no mutual recognition between these regimes. A single security incident could trigger multiple, distinct reporting timelines and formats, requiring organizations to maintain sophisticated internal processes to manage simultaneous, independent disclosure clocks without the benefit of a unified reporting standard.
For AI builders, the shift necessitates a move toward automated, audit-ready incident response. The financial stakes are substantial, with non-compliance under Article 14 carrying penalties of up to 15 million euros or 2.5% of global annual turnover, based on the Tier 1 penalty structure defined in Article 64. While full enforcement of the broader CRA obligations is slated for December 2027, the Article 14 reporting window is immediate. Initial enforcement efforts are expected to focus on the timeliness of the 24-hour filings and the accuracy of the root-cause analysis provided in the subsequent reports.
Article 14 establishes specific reporting timelines for AI products. By mandating a 24-hour clock for active exploits, the regulation requires AI manufacturers to adopt reporting processes similar to those used for critical infrastructure. While open-source software supplied on a non-commercial basis remains exempt, and web-only SaaS products may fall outside the CRA scope, the majority of commercial AI agent products are now firmly within the regulatory perimeter. Builders must now prioritize the development of robust detection and reporting pipelines to meet these requirements before the September 11 deadline.