Krux
April 9, 2026
90% Run AI Widely, 80% Miss Their Goals
Published: April 9, 2026 at 12:35 AM
Updated: April 9, 2026 at 12:35 AM
100-word summary
A survey of 124 CISOs reveals a troubling gap: nine in ten organizations report running generative AI widely, but only 20% of AI projects hit their performance targets. The culprit isn't the algorithms. Two-thirds of security chiefs say they can't confidently prevent unsafe access to AI training data, and 65% don't trust their own data security controls. The irony is sharp. Companies raced to put AI everywhere, then discovered their data was too messy, too poorly protected, or too chaotic to deliver results. The finding suggests most organizations built the AI engine before laying the tracks.
What happened
A survey of 124 CISOs reveals a troubling gap: nine in ten organizations report running generative AI widely, but only 20% of AI projects hit their performance targets. The culprit isn't the algorithms. Two-thirds of security chiefs say they can't confidently prevent unsafe access to AI training data, and 65% don't trust their own data security controls. The irony is sharp. Companies raced to put AI everywhere, then discovered their data was too messy, too poorly protected, or too chaotic to deliver results.
Why it matters
The finding suggests most organizations built the AI engine before laying the tracks.