Skip to main content

Accelerate AI use or ensure security?

Accelerate AI use or ensure security?

by João Paulo Spader Back, managing partner at SinapseTech

Anthropic published a figure in April this year that every technology executive should have read, and that few have taken seriously so far.

The company’s most recent model found and exploited, almost autonomously, serious flaws in systems used by billions of people. One of those vulnerabilities had existed for 27 years. The entire process cost less than US$ 50 and was completed in one night. The new model is about 90 times more effective than its previous version at turning known flaws into real attacks.

Given that, the debate many C-levels are still having internally, about accelerating AI adoption or prioritizing security, reveals a dangerous misunderstanding of the moment.

AI is already being used in your organization. By well-meaning employees, in unapproved tools, with real business data. What some call agility, the security market calls Shadow AI.

The risk does not come only from external actors. It also comes from within: integrations built without a security assessment, models accessing data that should never leave the company, decisions based on outputs nobody has audited.

AI governance does not mean blocking tools; it means creating the conditions for innovation to scale without creating invisible risk vectors. The question is not "when should we start governing AI use?". It is: when something goes wrong, will you be able to explain what happened?

If the answer is no, the time to act is now.