Survey: AI accidentally reached sensitive data at 87% of companies

Your AI might be going places it shouldn’t: Nearly nine in ten companies (87%) say an AI tool or agent got into sensitive data it didn’t need for its job in the past year, according to a new global survey by identity security firm Delinea. The firm polled more than 2.2k IT and security leaders in eight countries, including the UAE.

The good news: Companies have gotten the memo. Almost all of them (99.7%) now have written rules on what data AI can touch.

The bad news: Most of those rules exist only on paper, Delinea says. Only about half of companies check what their AI is doing against the rules as it happens — and fewer than one in five can catch an AI tool crossing the line in real time.

More bad news: Delinea also surveyed over 2.2k regular employees for the report. Nearly half (48%) said they always or regularly use AI to access company data without seeking formal approval.