Skip to content
Friday, October 23, 2026AboutContactRSS
EvidenceBook and Countersign Enable Tamper-Evident AI Agent Transaction Verification
Tech News

EvidenceBook and Countersign Enable Tamper-Evident AI Agent Transaction Verification

New software helps businesses independently verify records of automated interactions between artificial intelligence agents across corporate boundaries.

Key points

  • Action State Group Inc. launched open-source tools to address trust issues in automated inter-company transactions.
  • EvidenceBook provides tamper-evident records, while Countersign offers independent verification services for agent actions.
  • The system allows parties to compare signed transaction histories without requiring a central authority to hold data.

Action State Group Inc. has released a suite of open-source tools designed to allow businesses to independently verify the actions of artificial intelligence agents operating across corporate boundaries. According to SiliconANGLE, the offering addresses trust concerns as automated systems increasingly handle purchasing, selling and work delegation between organisations.

Root cause

The fundamental issue addressed by these tools is the lack of independent verification for automated interactions between different companies. Most existing observability solutions focus on internal systems within an organisation’s own infrastructure, leaving a gap for cross-border agent activities. Action State’s founder and chief executive Steven Mih noted that businesses often cannot rely solely on another party’s account of an agent’s actions when transactions occur externally.

Attack path

The tools do not prevent attacks but rather provide mechanisms to detect discrepancies in reported agent behaviour. EvidenceBook allows participants to maintain their own signed transaction histories, recording requests, payments and deliveries. By comparing these independent records, organisations can identify conflicts or gaps in the narrative without needing a central authority to store everyone’s data. Participants retain control over their information throughout this process.

Affected versions

The release includes three main components: EvidenceBook for preserving tamper-evident records, Terms Packs for evaluating agent behaviour and Countersign for independent verification. These tools utilise Agent Action Capsules, an open format for signed records, alongside cryptographic checkpoints to make subsequent changes detectable. A free public witness service establishes that a checkpoint existed by a specific time, ensuring temporal integrity of the evidence bundles.

Mitigation

  • Use EvidenceBook to maintain independent, signed records of agent interactions and transaction details.
  • Employ Terms Packs to evaluate whether agent behaviour aligns with agreed-upon parameters and expectations.
  • Utilise Countersign to check cryptographic properties of evidence bundles and record findings in a transparency log.
  • Compare selected records between parties to reveal agreements or conflicts without sharing full datasets.

Why it matters

Teams usually rely on internal logs to audit system behaviour, but these records are insufficient for verifying interactions with external automated agents. When AI agents operate across corporate boundaries, organisations need a way to independently confirm that reported actions actually occurred. This guidance suggests checking whether your organisation’s automated systems can produce tamper-evident records that can be compared against counterparties. Understanding the difference between recording an event and verifying its accuracy is crucial for maintaining trust in automated supply chains and inter-company workflows.

What to do and how to stay safe: Action State Group Inc.

  • Ensure your organisation maintains independent, cryptographically signed records of all automated agent interactions.
  • Verify that external partners can provide comparable evidence bundles for shared transactions or delegations.
  • Monitor for discrepancies between internal logs and external agent reports during routine audits.
  • Review agent behaviour against established terms packs to identify deviations from agreed protocols.

Step-by-step guide: Password Policies Examples That Reduce Password Fatigue

General security guidance from the Vector Bulletin newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Does EvidenceBook guarantee that an agent’s actions were accurate or complete?

No, Mih stated that preserving a record does not establish content accuracy or that every action was captured.

Can Countersign certify that a claimed outcome actually occurred?

No, the service checks cryptographic properties and signs findings but does not certify claimed outcomes or evidence completeness.

Do both parties need to install the software to benefit from verification?

No, one party can benefit without the other installing the software, though compatible implementations allow evidence exchange.

Sources

  1. SiliconANGLE
Action State Group Inc.EvidenceBookCountersignAI agentsSiliconANGLE

Related stories

1Password Introduces Agent Login System That Hides Credentials from AI

A new platform allows AI assistants to sign into accounts without ever viewing or storing user passwords, requiring explicit approval for each access request.