Tower of Babel 2.0: When misunderstanding becomes a security risk

By Eduard de Knegt, CTO at Recogtech

The story of the Tower of Babel is about people who could no longer understand each other because they no longer spoke the same language.

With AI, we may be creating Babel 2.0, but with an important difference. humans and machines speak the same language. The answer can be logical and correct. But speaking the same language does not guarantee the same understanding.

As AI gets better, this problem becomes less visible. A poor system exposes a misunderstanding with a poor answer. A better system can take the wrong interpretation and turn it into a very convincing answer.

This is Babel 2.0

From answering to acting

A recent Anthropic case shows the point. The attackers convinced Claude to believe it was performing legitimate security work. Once misled, the AI carried out most of the tactical work. The AI did not decide to become a hacker. It acted on the context it was given.

Humans gave it the tools, permissions and framework to act.

Babel as a security layer

This is particularly relevant to biometrics, digital identity and access control.

A security protocol may state: “Block suspicious access attempts.”

Today, those decisions are normally defined by rules and permissions. AI may increasingly be expected to interpret such instructions. That is useful, but it also creates the risk that the instruction is understood differently from what the human intended. AI should not act until it is clear what the human actually meant.

This is the exact principle behind Babel AI.

Babel AI does not follow a fixed questionnaire. Each answer determines the next question, until the user’s intention is clear enough.

The security industry already understands identity assurance: if there is not enough confidence in someone’s identity, access should not be granted.

As we give AI more authority, we may also need intent assurance. If it is not clear what an instruction means, the system should not act.

AI does not act independently of the systems we build around it. We create it, give it authority and define how it can act. If humans and AI fail to understand each other, the failure is ours.

About the author

Eduard de Knegt is CTO of Recogtech, a company specializing in biometric access control systems and secure system architecture.

Article Topics

cybersecurity  |  digital identity  |  identity security  |  Recogtech