An analysis of data collected by Netskope published today finds the number of individuals accessing generative artificial intelligence (AI) applications via a cloud service increased threefold in the past year, with the number of prompts sent increasing sixfold, from 3,000 per month on average to 18,000.

A third of organizations (33%) are using OpenAI services via the Microsoft Azure cloud, compared to 27% using the Amazon Bedrock service and 10% using the Google Vertex AI. A full 70% connect to api.openai.com, 54% to AssemblyAI, and 30% to Anthropic application programming interfaces (APIs).

At the same time, the report also noted there has been a major drop in the number of end users relying on personal accounts to access these services. Personal AI app usage dropped from 78% to 47%, while usage of managed accounts provided by organizations increased from 25% to 62%, according to the report. However, the report also notes that end users are now also sharing more personal data with these services via the accounts they created.

Ray Canzanese, director of Netskope Threat Labs, said that while increased reliance on managed corporate accounts should provide cybersecurity teams with more visibility into how these services are employed, risks remained high in the absence of any consistent approach to governance and security. It remains, for example, relatively trivial to create a prompt injection attack through which adversaries can now compromise an entire workflow, he noted.

As such, IT and cybersecurity teams need to gain more visibility into how these services are being employed to ensure policies are being followed, said Canzanese. Overall, the report finds GenAI data policy violation incidents doubled in the past year, with the average organization seeing 223 incidents per month. ​Most of those violations appear to involve a relatively narrow group of end users, with just 3% of genAI users in the average organization committing data policy violations.

The report also finds half of organizations (50%) lack enforceable data protection policies for generative AI apps, with the most incidents involving source code (42%), regulated data (32%) and intellectual property (16%).

The challenge is that as generative AI applications are being added to environments, other attack vectors are still being exploited. For example, cybercriminals continue to successfully use trusted platforms to distribute malware, noted Canzanese. GitHub is the most abused service for malware distribution, with 12% of organizations detecting employee exposure monthly, followed by Microsoft OneDrive (10%) and Google Drive (5.8%), according to the report.

In general, the number of cybersecurity incidents traced back to third-party services has shown a spotlight on how vulnerable interconnected systems have become, noted Canzanese.

On the plus side, however, the number of individuals clicking on phishing links declined year-over year, with now only 87 out of every 10,000 users clicking on a phishing link each month.

It’s not clear to what degree cybersecurity teams might soon be overwhelmed in the age of AI, but the need for increased visibility into how various services are being used has become crucial. The challenge, as always, is to ensure the overall IT environment remains secure as it continues to dynamically evolve and expand.