SpyCloud forecasts stolen AI credentials and trusted connections as 2027 risks
SpyCloud has released its 2026 Forecast Report, which sets out the company’s predictions for cyber threats in 2027. The announcement identifies two areas of concern: stolen credentials associated with artificial intelligence systems and trusted connections that attackers could exploit.
The report’s predictions are described as drawing on expertise from SpyCloud’s security research, product, investigations and data science teams. The announcement does not provide the report’s detailed findings, examples, affected products, attack methods or recommended controls. It also does not quantify the expected scale or likelihood of the risks. Those limits matter: the headline points to concerns for the coming year, not evidence that a specific campaign or incident is underway.
Why identity teams should pay attention
Credentials can provide access under a legitimate identity, making their theft relevant to customer identity and access management as well as workforce security. In an AI context, organizations may need to understand which accounts, services and integrations can interact with AI tools, and what data or actions those identities are authorized to reach. The announcement does not specify which kinds of AI credentials SpyCloud means, so organizations should not assume it refers to one particular product, credential type or compromise route.
The reference to trusted connections raises a related question for defenders: whether an account or integration that is accepted as legitimate could be misused after credentials are exposed. That is a general identity-security consideration, rather than a description of a particular attack chain in the announcement. Reviewing authorization, limiting access to what each identity needs, and ensuring that credentials and integrations can be revoked are practical areas for organizations to assess.
Use the forecast as a prompt for review
Security and IT teams can use the announcement as a reason to inventory identities used by AI services and connected applications. That review can establish who owns each account or integration, what systems it can access, whether its permissions remain necessary, and how access is monitored. Teams should also confirm that their incident processes cover suspected credential exposure, including the ability to disable or rotate credentials and investigate activity associated with the affected identity.
These are general defensive steps, not measures that SpyCloud is reported here to have prescribed. The available announcement does not include specific indicators of compromise, technical guidance, affected versions, or a response checklist. Organizations seeking the forecast’s full analysis would need to consult the report itself; its detailed recommendations cannot be inferred from the brief release notice.
For enterprise leaders, the immediate takeaway is to treat AI-related identities and connections as part of the broader identity governance picture. The forecast flags them as issues to watch in 2027, while the information provided does not establish how attackers may target them or how frequently such activity will occur. A measured review of access and ownership can help teams prepare without treating a prediction as confirmation of an active threat.