As AI agents move from answering questions to initiating transactions and interacting with customers, proving that an action was actually authorized by a human is becoming a new security challenge. ValidSoft is addressing that gap with Cryptographic Intent Binding, an architecture designed to link a verified human identity to the specific instruction or transaction that person authorized.

ValidSoft Wants to Add a Missing Layer to Agentic AI Security: Proof of Intent

Identity has traditionally been the starting point for digital security. Authenticate the person, confirm the credentials and allow the transaction.

Agentic AI complicates that model.

An AI agent can increasingly interpret instructions, interact with systems and execute actions on behalf of users. That creates a security question that conventional authentication does not necessarily answer: Did the authenticated person actually authorize this specific action?

ValidSoft says its new Cryptographic Intent Binding architecture is designed to provide that missing layer.

The UK-based security technology company has announced technical details of an approach intended to cryptographically associate a verified human identity with the precise content of an authorized instruction. The technology is particularly aimed at environments where AI agents, voice interfaces and automated transactions could create ambiguity around who authorized an action and what they actually approved.

Use up and down arrow keys to resize the meta box pane.

The distinction between identity and intent could become increasingly important as enterprises deploy AI agents into higher-risk workflows.

Authentication Can Prove Who You Are. Intent Is Different.

Traditional authentication answers a relatively straightforward question: Who is attempting to access the system?

Biometrics, passwords, multifactor authentication and device credentials can all contribute to that determination.

But authentication does not necessarily establish what the user intended to authorize.

Consider an AI-powered financial service. A customer might authenticate successfully and tell an AI agent to perform a transaction. If the agent interprets the instruction incorrectly, or if an attacker manipulates the interaction, proving the user’s identity alone may not establish whether the final action represented the customer’s actual intent.

That distinction becomes more consequential as AI agents gain greater autonomy.

ValidSoft’s architecture is designed around the proposition that enterprises need a cryptographically verifiable record connecting identity, authorization and action.

VoiceMFA Creates a Cryptographic Link Between Person and Instruction

At the center of ValidSoft’s approach is VoiceMFA™ (See-Say).

The technology uses cryptographic hashing and device-bound codes to associate a verified voice identity with the content of a particular authorized instruction or transaction.

According to ValidSoft, the resulting authorization record is designed to provide four characteristics:

Genuinely authorized: The action is explicitly approved by the person whose authority is being used rather than inferred from authentication alone.
Provable: The authorization can be demonstrated later to auditors, regulators or other third parties.
Non-repudiable: The authorization provides evidence that can make subsequent denial of the captured intent more difficult.
Immutable: The authorization record is designed to resist alteration or retrospective modification.

For security teams, the important development is the move from an authentication event to an evidence-bearing authorization event.

That could matter in regulated environments where organizations need to demonstrate not merely that a customer accessed an account, but that a specific transaction was properly authorized.

Three Layers of Voice Security

ValidSoft plans to deploy VoiceMFA within its AI Voice Identity Platform (VIP™), alongside two other capabilities.

VoiceID™ provides voice biometric identity verification, establishing who the speaker is.

Voice Verity® is designed to detect deepfake and synthetic speech, addressing the growing threat of AI-generated or manipulated voices.

VoiceMFA™ (See-Say) adds the authorization layer, associating the verified identity with the specific action being approved.

The architecture effectively separates three questions:

Is this a real person?

Is this the person they claim to be?

What exactly did that person authorize?

That layered model reflects a broader change in security architecture as generative AI and synthetic media make voice-based authentication more difficult to trust in isolation.

Agentic AI Raises the Stakes

The timing of the announcement is closely tied to the development of agentic AI.

Generative AI systems initially operated primarily as conversational interfaces. Increasingly, AI agents are being designed to interact with enterprise systems, retrieve information, initiate workflows and execute actions on behalf of users.

That creates an expanded attack surface.

An attacker no longer necessarily needs to steal a password if they can manipulate an AI agent, compromise an authenticated session or exploit ambiguity between a user’s request and an agent’s interpretation.

This is one reason the security industry is moving toward concepts such as zero-trust architecture, continuous authentication, transaction signing and cryptographic authorization.

ValidSoft’s approach extends that thinking into voice and AI-agent interactions.

Rather than treating authentication as the end of the security process, Cryptographic Intent Binding treats authorization itself as an event that should be independently captured and protected.

Contact Centers Could Be an Early Use Case

Contact centers are another area where the technology could have practical implications.

Voice remains an important interface for banking, insurance, healthcare, telecommunications and other customer-service environments. At the same time, these organizations increasingly use AI to automate conversations and support agents.

The challenge is that voice interactions are inherently difficult to audit at the level of intent.

A recorded conversation can establish what was said, but it does not automatically provide a cryptographically protected proof that a particular person authorized a particular action.

ValidSoft says its framework is designed for contact centers, voice and agentic payment channels, as well as compliance and audit workflows.

For enterprises, this could make intent evidence part of transaction security rather than an after-the-fact analysis exercise.

Security Vendors Are Moving Beyond Detection

The larger market is shifting from simply detecting threats toward proving the legitimacy of actions.

Companies such as Microsoft, Google, NVIDIA, Okta and other cybersecurity and identity providers are developing security architectures around AI agents, identity and increasingly autonomous software.

The emerging challenge is that agentic systems blur the boundary between human instruction and machine execution.

An enterprise may authenticate a human, authorize an AI agent and then allow that agent to execute multiple downstream actions. Each step introduces another opportunity for ambiguity.

Cryptographic Intent Binding attempts to address that problem by creating a stronger relationship between the original human authorization and the action that follows.

Whether that model becomes a mainstream security layer will depend on interoperability, deployment complexity, regulatory acceptance and how effectively the technology works against increasingly sophisticated voice manipulation.

But the underlying problem is unlikely to disappear.

As AI agents gain the ability to act rather than simply respond, proving what a human intended may become as important as proving who the human was.

Market Landscape

Agentic AI is forcing enterprises to rethink conventional identity and access management.

Traditional authentication systems are largely designed around establishing a user’s identity and granting access according to predefined permissions. Agentic AI introduces another dimension: authorization of actions performed by software on a user’s behalf.

That creates potential demand for technologies spanning:

AI agent identity and authentication
Transaction-level authorization
Voice biometrics
Deepfake and synthetic speech detection
Cryptographic signing
Zero-trust security
Digital identity
Audit and compliance infrastructure

The competitive landscape includes identity platforms from Microsoft, Okta and Google, cybersecurity vendors developing AI-agent protections, and specialized companies working on voice authentication and synthetic-media detection.

ValidSoft’s differentiation is its attempt to combine voice identity, deepfake detection and cryptographically bound intent into one security architecture.

For financial institutions and other regulated enterprises, the question will be whether these technologies can become interoperable components of existing identity, fraud and transaction-security stacks rather than another isolated security layer.

Top Insights

ValidSoft’s Cryptographic Intent Binding targets agentic AI security, linking verified human identity with specific instructions rather than relying solely on conventional authentication.
VoiceMFA™ combines voice identity with cryptographic authorization, creating an intended audit trail for transactions and AI-agent actions in sensitive environments.
ValidSoft’s platform combines detection, identity and intent, using Voice Verity, VoiceID and VoiceMFA to create layered voice security.
Contact centers and payments could benefit most, particularly where organizations need stronger evidence that customers explicitly authorized automated transactions.
Agentic AI is expanding the identity-security problem, making proof of human intent increasingly important as software gains authority to act independently.

Get in touch with our Adtech experts