Abstract

A digital signature can be valid, an AI agent can be authenticated, and an API request can pass every security check—yet the resulting action may still be unauthorized.

Modern software increasingly performs actions without direct human supervision. Artificial intelligence (AI) agents can initiate payments, transfer digital assets, send information, execute code, modify cloud infrastructure, access protected services, and control connected devices. These operations often rely on established security mechanisms, including TLS, OAuth 2.0, digital signatures, API credentials, access-control policies, hardware security modules (HSMs), and trusted execution environments.

However, successful authentication does not necessarily establish that the exact operation about to be executed is still authorized.

For example, an AI agent may receive permission to transfer $100 but prepare a transaction for $1,000. An authenticated service may send confidential data to an unintended destination. A valid digital signature may authorize installation of unexpected executable code. A cloud automation system may modify a protected resource after its authorization has expired. A connected device may execute a command that no longer satisfies current operating restrictions.

In these situations, the identity, credential, signature, or communication channel may be valid even though the actual operation differs from the originally approved action.

This technical disclosure describes Protected Transaction Execution Finality (PTEF), an experimental, application-layer security architecture that introduces independent authorization verification at the point where a digital operation becomes effective.

The proposed system separates the authority to request an operation from the authority to execute it. An independently protected component records the originally approved action, including the authorized identity, operation type, destination, permitted parameters, execution environment, validity period, policy version, and usage restrictions.

The proposed operation is initially held in a non-executable or otherwise non-effective state. Immediately before execution, a separate protected verification component independently reconstructs the actual pending operation from its final command, transaction data, executable code identity, destination, or device state.

The system compares the reconstructed operation against the original authorization and verifies current policy, revocation status, execution context, spending or resource limits, replay protection, and authorization validity. Only after the required checks succeed may an operation-specific, single-use or bounded authorization be consumed at the protected execution boundary.

The framework is designed to prevent an untrusted AI model, workflow engine, compromised service, or intermediary component from independently converting a valid credential into unrestricted permission to perform a consequential operation.

The architecture also addresses authorization changes during processing, time-of-check/time-of-use vulnerabilities, transaction replay, delegated-code changes, concurrent resource consumption, cross-platform execution, failure recovery, and alternative execution paths that might otherwise bypass security enforcement.

Applications include AI-agent tool execution, autonomous financial transactions, digital-asset custody, smart-contract delegation, banking APIs, cloud administration, secure data export, software sandboxing, telecommunications, connected devices, industrial control, and cryptographic signing infrastructure.

The 213-page runnable technical disclosure provides 59 security and implementation profiles, 148 reproducible test-vector bundles, complete reference implementations in Python, Go, and JavaScript, detailed execution workflows, state-transition rules, security invariants, and graphical decision diagrams.

The accompanying laboratory results report 444 successful cross-language checks out of 444, with zero reported failures across the three reference implementations. The runnable PDF edition also includes source files, SHA-256 integrity fingerprints, automated test-generation tools, and measured local execution latency.

In one documented laboratory environment, the median in-process authorization decision for the full verification-and-consumption scenario was approximately 5.7 microseconds in Go, 10.2 microseconds in Node.js, and 12.9 microseconds in Python. These measurements concern the educational decision engines and exclude production storage, hardware signing, network transmission, and external settlement.

The reference implementations demonstrate consistent local authorization logic across programming languages. They do not establish production cryptographic assurance, independent security certification, complete enforcement-path coverage, or protection against all possible threats.

PTEF is intended to complement existing authentication, cryptographic signing, identity management, network security, and trusted-computing technologies rather than replace them.

Its central security principle is:

A system should verify what an operation will actually do, whether that exact operation remains authorized, and whether permission can be safely consumed—before allowing the operation to take effect.

The disclosure is associated with Indian Provisional Patent Application No. 202631130246. Earlier technical disclosures include the author's principal international patent application PCT/IB2026/055615 (WO 2026/150382), concerning the broader execution-finality architecture. The precise priority and subject-matter relationships between applications are governed by their respective filed documents.

This publication is intended for patent examiners, patent attorneys, AI infrastructure developers, cloud providers, cybersecurity engineers, financial-technology companies, hardware security developers, telecommunications engineers, standards researchers, and organizations developing autonomous digital systems.

International Patent Classification (IPC) — Proposed Search Codes

Primary IPC classifications

  • G06F 21/30 — Authentication and authorization of computer-system security principals.

  • G06F 21/53 — Security through restricted execution environments, including sandboxes and secure virtual machines.

  • G06F 21/62 — Protection of data access using platform-based security and access-control mechanisms.

  • H04L 9/32 — Cryptographic authentication and verification of identity or authority.

  • H04L 63/10 — Control of access to network resources and devices.

Additional IPC classifications

  • G06F 21/57 — Maintaining trustworthy computer platforms, including firmware and software integrity.

  • G06F 21/64 — Protection of data integrity.

  • G06Q 20/40 — Authorization of electronic payment transactions.

  • G06Q 20/38 — Electronic payment protocols.

  • G05B 19/042 — Digital-processor-based programmable control.

  • G05B 19/048 — Safety and monitoring in programmable control systems.

Cooperative Patent Classification (CPC) — Proposed Search Codes

Primary CPC classifications

  • G06F 21/30 — Authentication and authorization of security principals.

  • G06F 21/52 — Computer security during program execution.

  • G06F 21/53 — Restricted execution environments and sandbox security.

  • G06F 21/62 — Computer-platform access control.

  • H04L 63/10 — Network-resource access control.

  • H04L 9/32 — Cryptographic authentication and authority verification.

Additional CPC classifications

  • G06F 21/57 — Trusted computing and platform-integrity maintenance.

  • G06F 21/64 — Data-integrity protection.

  • G06F 21/554 — Local security event detection and direct countermeasures.

  • H04L 63/08 — Authentication of network entities.

  • H04L 63/123 — Verification of received data contents.

  • H04L 9/3228 — Cryptographic authentication or authorization using one-time or temporary information.

  • G06Q 20/40 — Payment authorization.

  • G05B 19/0428 — Safety and monitoring functions in processor-based control systems.

Classification notice: These IPC and CPC codes are proposed search classifications reflecting different technical implementations disclosed in the document. They are not official classifications assigned to Indian Patent Application No. 202631130246, and not every code will necessarily apply to each patent claim.

Keywords for Patent Examiners, AI Infrastructure and Industry Discovery

AI agent security, agentic AI, autonomous agents, AI tool execution, AI agent authorization, AI agent sandbox security, AI runtime security, autonomous workflow security, execution-time authorization, protected execution, execution finality, protected transaction execution finality, authorization-to-execution consistency, intent-to-execution verification, intent-to-effect binding, action-specific authorization, cryptographically valid unauthorized operation, digital signature verification, cryptographic signing policy, hardware security module, HSM authorization, secure key usage, trusted execution environment, TEE, confidential computing, protected runtime enforcement, privilege separation, zero trust security, OAuth 2.0, API authorization, API gateway security, service mesh security, privileged API execution, cloud infrastructure security, cloud resource modification, transaction integrity, digital asset custody, cryptocurrency wallet security, EIP-7702 account delegation, smart contract authorization, digital payments, payment execution security, replay attack prevention, nonce reuse prevention, one-time authorization, non-bearer cryptographic capability, policy revocation, policy version enforcement, atomic authorization consumption, idempotent transaction execution, distributed quota management, race-condition protection, TOCTOU prevention, crash-safe transaction processing, execution path mediation, non-bypassable security enforcement, secure data export, data exfiltration prevention, secure network transmission, device command authorization, industrial control security, hardware-assisted execution security, remote attestation, firmware integrity, cross-domain security interoperability, autonomous system safety.

Additional Patent Examiner Search Expressions

"AI agent" AND "execution-time authorization"

"autonomous agent" AND "tool execution security"

"digital signature" AND "operation authorization"

"cryptographic signing" AND "execution policy"

"authorization" AND "actual operation verification"

"secure execution" AND "protected authorization"

"AI agent sandbox" AND "network egress control"

"API gateway" AND "non-bypassable enforcement"

"trusted execution environment" AND "policy verification"

"hardware security module" AND "transaction authorization"

"single-use authorization" AND "atomic consumption"

"capability-based security" AND "execution authorization"

"delegated code" AND "authorization verification"

"smart contract" AND "execution-time authorization"

"time-of-check time-of-use" AND "transaction integrity"

"policy revocation" AND "in-flight operations"

"distributed systems" AND "replay prevention"

"cloud infrastructure" AND "privileged action authorization"

"AI agent" AND "data exfiltration prevention"

"device actuation" AND "execution authorization"

"cross-domain authorization" AND "interoperability"

"transaction execution" AND "protected state"

"cryptographically valid" AND "unauthorized operation"

Patent Application and International Technical Disclosure References

Present Indian patent application

  • Indian Provisional Patent Application No.: 202631130246

  • Jurisdiction: India

  • Associated subject: Protected transaction authority and execution-finality mechanisms for delegated digital operations

  • Related international PCT filing: Awaiting filing, as indicated by the author; number not supplied

Principal earlier international application — Mothership

  • International Application No.: PCT/IB2026/055615

  • Filing Date: 4 June 2026

  • International Publication: WO 2026/150382

  • Technical subject: Secure execution-finality architecture, protected authorization, execution-boundary verification, and cross-domain autonomous digital operations.

  • Author-reported specification length: Approximately 8,598 pages.

Other previously identified international applications addressing related technical areas

  • PCT/IB2025/058316 — Earlier ALF-related international application.

  • PCT/IB2026/055760 — DAS Protocols Part II.

  • PCT/IB2026/055870 — DAS Protocols Part III.

  • PCT/IB2026/056058 — DAS Protocols Part IV.

  • PCT/IB2026/056571 — DAS Protocols Part V.

  • PCT/IB2026/056941 — DAS Protocols Part VI.

  • PCT/IB2026/057198 — DAS Protocols Part VII.

  • PCT/IB2026/058236 — DAS Protocols Part IX.

These earlier application numbers are cross-references for technical discovery and do not establish that all applications belong to one legally recognized patent family, or that a feature in the present disclosure necessarily has the priority date of an earlier application.

Reference Implementation and Reproducibility

The Runnable Edition includes a directly inspectable experimental reference implementation and supporting test materials.

  • Implementation languages: Python, Go, JavaScript (Node.js).

  • Test-vector bundles: 148.

  • Reported cross-language checks: 444.

  • Reported successful checks: 444.

  • Reported failed checks: 0.

  • Test coverage: Authorization matching, altered transaction fields, context mismatch, delegation restrictions, policy revocation, replay, quota enforcement, malformed input, missing authority, unknown instruction origin, and related state-security conditions.

  • Reproducibility materials: Embedded runnable source files, test runner, vector generator, SHA-256 fingerprints, latency harness, and accompanying instructions.

The test results represent documented laboratory execution of the supplied reference models. They are not independent formal verification or evidence of production security certification.

Publication, Patent and Technical Status

Publication status: Public technical disclosure and experimental proposal.

Related IETF document: draft-das-protected-transaction-execution-finality-00. The Internet-Draft is a proposal for technical discussion and is not an adopted IETF standard.

Patent notice: Indian Provisional Patent Application No. 202631130246. Patent pending. Any patent rights and their scope depend on the applicable filed specifications, claims, national laws, and subsequent examination.

Copyright license stated in the uploaded edition: Creative Commons Attribution 4.0 International (CC BY 4.0). Patent rights are separately reserved; the copyright license does not itself grant patent rights.

Technical limitation: The framework is enforceable only at execution paths actually controlled by the protected security mechanism. Unrestricted signing keys, uncontrolled alternate execution routes, compromised trusted components, and previously released effective authorizations may fall outside its protection.

Central technical distinction:

Authentication asks: Who is requesting this action?

Authorization asks: What is that identity permitted to do?

Protected execution-time verification asks: Is this exact action, with its current parameters and consequences, still permitted at the moment it becomes effective?

More details :,

Author: Sangam Kumar Das Affiliation: Independent Inventor and Researcher, India Indian Provisional Patent Application No.: 202631130246 Related PCT Filing: Awaiting filing; no corresponding PCT application number identified Earlier Principal International Patent Application (Mothership): PCT/IB2026/055615 International Publication: WO 2026/150382 Technical Disclosure Date: 1 November 2026 Publication Type: Technical Disclosure — Runnable Reference Implementation Edition Related IETF Internet-Draft: draft-das-protected-transaction-execution-finality-00 (Experimental)

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License

Download

Share

COinS