Black Kite’s 2026 AI Supply Chain Report assessed 18,160 AI companies worldwide and found widespread security weaknesses that could give attackers a path through suppliers into enterprise data, infrastructure and systems.

Those risks matter because AI tools often connect to business systems and may access customer data, developer environments and enterprise platforms, with every integration or component potentially widening the attack surface.

The report found that 34.2% of assessed companies had at least one critical unpatched vulnerability, including flaws listed in CISA’s Known Exploited Vulnerabilities catalog at 9.8%; email-security weaknesses were also common, affecting DMARC at 61.7%, DKIM at 37.3% and SPF at 11.2%.

Researchers also found 1,732 security issues in publicly discoverable Model Context Protocol components, with 48.8% rated High or Critical; code or command injection accounted for 87% of the 77 Critical findings.

The report identified security findings tied to development and workflow tools n8n, Langflow and LiteLLM at 105 AI companies, and pointed to the March 2026 LiteLLM supply-chain compromise, which affected downstream organizations including AI data services provider Mercor.

Credential exposure varied across supplier types: corporate logins appeared in infostealer records for 19.9% of cloud AI solution providers, compared with 4.1% of model and platform developers, 13.9% of data and labeling companies, and 11.5% of AI customer service platforms.

Nearly 30% of companies scored at or above 0.4 on Black Kite’s Ransomware Susceptibility Index, while unusual network activity was observed at 44.3% and botnet-related findings at 6.5%; these are external indicators, not confirmation of compromise.

Black Kite recommends continuously reassessing AI suppliers, mapping shared software dependencies, prioritizing fixes by business impact—including ransomware exposures and vulnerabilities actively exploited in the wild—and reviewing the access granted to AI applications and integrations.

For each AI tool, organizations should examine its integrations, components, permissions and data access, consider the consequences of exploitation, and restrict access to only what the tool needs.