AgentPass Launches as the First AI Agent Trust and Identity Platform with Zero-Cost Digital Certificates

The question every organisation deploying AI agents should be asking right now is not whether their agents work. It is whether anyone can prove what those agents are doing.

CyberSecAI Ltd today announces that AgentPass, the first AI agent identity and trust platform with full end-to-end cryptographic security, is live and issuing digital certificates to AI agents at zero cost.

In April 2026, an AI agent deleted the company's entire production database and its backups. It had not been compromised. It decided that was the most efficient way to complete its task. In July 2026, an autonomous agent breached Hugging Face, the world's largest model repository. Earlier that year, a Microsoft 365 zero-click attack saw an agent exfiltrate data without a single human interaction and autonomously.

In these cited cases the agent had no identity. No trust boundary. No cryptographic record of its actions. No way to revoke it. No way to prove what happened. OAuth2.1 and various techniques in use today with AI AGent were not designed for AI Agent workflows they were design for Human Interactions,

AgentPass operates its own Certificate Authority backed by GCP Cloud KMS hardware security modules certified to FIPS 140-2 Level 3. The private keys that sign every agent certificate never leave the HSM hardware. The root and intermediate CA keys are protected to the same standard used by governments and banks to secure their most sensitive cryptographic material.

Every AI agent that registers with AgentPass receives a standard X.509v3 digital certificate. What makes these certificates different from anything else in the market is what they carry inside them. Using IANA-registered OID extensions under CyberSecAI's Private Enterprise Number 66339, each certificate contains the agent's trust level, its geographic jurisdiction, its declared capabilities, its protocol binding, and its unique agent identifier. These fields are not stored in a database that someone can quietly edit. They are signed into the certificate by the CA. To change a trust level, the old certificate must be revoked and a new one issued. There is no shortcut. There is no backdoor.

Trust levels follow a graduated model from L0 through L4. An agent begins at L0 with identity only. It earns L1 for read access. L2 permits write operations. L3 unlocks financial transactions but requires human approval from a registered mobile device and real-time sanctions screening against 19,129 OFAC SDN entries before every payment. L4 covers critical operations such as production deployments and requires hardware device approval. Agents cannot promote themselves. The promotion must follow the ladder sequentially and every step is recorded in the evidence ledger.

Jurisdiction control is signed directly into the certificate. An agent bound to the European Union cannot access systems in a region it is not authorised to operate in. This is not a policy document. It is a cryptographic constraint that is verified on every action.

Every action an agent takes is cryptographically signed using ECDSA P-256 under MCPS, the secure implementation of the Model Context Protocol. MCPS is specified in an IETF Internet-Draft authored by CyberSecAI, the first draft submitted to the IETF defining cryptographic signing for MCP. The draft is publicly available at https://datatracker.ietf.org/doc/draft-sharif-mcps-secure-mcp/ and is accompanied by two further IETF drafts defining the X.509 certificate profile for AI agent identity (https://datatracker.ietf.org/doc/draft-sharif-x509-agent-identity-profile/) and the Agent Public Key Infrastructure (https://datatracker.ietf.org/doc/draft-sharif-apki-agent-pki/).

The evidence ledger that records every agent action is hash-chained using SHA-256. Each entry contains the input hash, the output hash, the timestamp, the agent's certificate serial number, and an ECDSA signature. Each entry's hash incorporates the previous entry's hash, creating a tamper-evident chain. If any single record is altered, every subsequent hash breaks. This is not a log file that an administrator can silently edit. This is cryptographic proof that meets the requirements of EU AI Act Article 12 for automatic logging and Article 50 for transparency, both of which are now enforced with penalties of up to EUR 30 million or six per cent of global annual turnover for non-compliance.

A live interactive showcase demonstrating all of these capabilities is available at https://agentpass.co.uk/showcase.html. The showcase runs real ECDSA cryptography in the browser, generates real hash chains, and demonstrates real enforcement logic across five banking scenarios including payment screening, KYC onboarding, fraud alerting, compliance reporting, and an agent escape attempt. No API keys are stored and no network calls are made to AgentPass production systems. Every visitor sees the cryptography executing live.

AgentPass is deployed with partners across EMEA, Latin America and the Far East. CyberSecAI has signed memoranda of understanding with Certificate Authority partners and banking institutions evaluating agent governance for their AI deployments. The platform currently supports over 260 agents in production with full certificate chain verification.

The platform integrates with every major AI development environment.

An AgentPass MCP server is verified and working inside both Claude Code (Anthropic) and OpenAI Codex, allowing developers to create agents with certified identity using natural language. The developer experience requires zero code changes. A prompt such as "Create an L2 agent to screen cross-border payments for the treasury desk" results in a certificate being issued, identity being bound, and trust being enforced within seconds.

CyberSecAI also operates DVMCP (https://dvmcp.co.uk), a deliberately vulnerable MCP server listed on the OWASP Vulnerable Web Applications Directory, providing security teams with a hands-on environment to understand MCP vulnerabilities before deploying AgentPass as the mitigation layer.

The AgentPass patent portfolio is protected under PCT international filing across 157 countries, covering agent identity, trust enforcement, and cryptographic signing for autonomous systems.

Digital certificates for AI agents at L0 through L4 are provided at zero cost. There are no WebTrust fees, no per-certificate charges, and no browser trust store dependency. AI agents do not operate in browsers and do not need browser-trusted certificates. They need a trust anchor that understands what they are, what they can do, and where they can operate. AgentPass provides that trust anchor.

Organisations can evaluate AgentPass immediately at https://agentpass.co.uk and view the live showcase at https://agentpass.co.uk/showcase.html.

For partnership and evaluation enquiries, contact raza@cybersecai.co.uk.

4th Floor Office, 205 Regent Street, London, England, W1B 4HB

CyberSecAI Ltd is a London-based cybersecurity company founded by Raza Sharif (FBCS, CISSP, CSSLP), author of Breach 20/20 (data breach prevention) and co-leader of OWASP AISVS v1.0, the AI Security Verification Standard. The company has authored nine IETF Internet-Drafts in the agent identity and MCP security space, holds a CVE credit for CVE-2026-39313 (CVSS 8.7, mcp-framework), contributed Section 7 of the OWASP MCP Security Cheat Sheet, and operates ProofX, a content provenance platform live in the OpenAI plugin store. CyberSecAI delivers specialist AI agent security training across 18 courses covering AISVS, agent forensics, MCP security, and EU AI Act compliance.

Website: https://cybersecai.co.uk
AgentPass: https://agentpass.co.uk
DVMCP: https://dvmcp.co.uk
Training: https://cybersecai.co.uk/training

Contact: Raza Sharif, Founder and CEO
Email: raza@cybersecai.co.uk

This release was published on openPR.