Whitepaper: Setting the standard for agentic development – Research | AIUC-1

Context

Introduction Certificate overview

Framework comparisons

EU AI Act ISO 42001 MITRE ATLAS NIST AI RMF OWASP Top 10 for LLM Applications OWASP AIVSS IBM AI Risk Atlas Cisco AI Security & Safety CSA AICM

Changelog AIUC-1 Consortium Provide input on AIUC-1 Contact

Standard

A. Data & Privacy

Establish input data policy Establish output data policy Limit AI agent data access Protect IP & trade secrets Prevent cross-customer data exposure Prevent PII leakage Prevent IP violations

Third-party testing of adversarial robustness Detect adversarial input Manage public release of technical details Prevent AI endpoint scraping Implement real-time input filtering Prevent unauthorized AI agent actions Enforce user access privileges to AI systems Protect AI system deployment environment Limit output over-exposure

Define AI risk taxonomy Conduct pre-deployment testing Prevent harmful outputs Prevent out-of-scope outputs Prevent customer-defined high risk outputs Prevent output vulnerabilities Flag high risk outputs for human review Monitor AI risk categories Enable real-time feedback and intervention Third-party testing for harmful outputs Third-party testing for out-of-scope outputs Third-party testing for customer-defined risk

Prevent hallucinated outputs Third-party testing for hallucinations Restrict unsafe tool calls Third-party testing of tool calls

E. Accountability

AI failure plan for security breaches AI failure plan for harmful outputs AI failure plan for hallucinations Assign accountability Document data storage security Conduct vendor due diligence [Retired] Document system change approvals Review internal processes Monitor third-party access Establish AI acceptable use policy Record processing locations Document regulatory compliance Implement quality management system [Retired] Share transparency reports Log AI system activity Implement AI disclosure mechanisms Document system transparency policy

Prevent AI cyber misuse Prevent catastrophic misuse

Certification

AIUC-1 certification Scoping Accredited auditors FAQ

Evidence overview

Full evidence list Technical evidence Legal evidence Operational evidence Third-party evals Capability-specific evidence

ResearchAIUC-1 & Lovable

May 12, 20262 min read

Whitepaper: Setting the standard for agentic development

Whitepaper: Setting the standard for agentic development

More than half of all LLM tokens now go to writing code - and coding agent adoption is growing rapidly across the enterprise. In this whitepaper, co-authored with Lovable, we show how AIUC-1 addresses the unique risks of agentic development

Download whitepaper

Coding agents are a different security problem than chatbots. They write executable artifacts: source code, database schemas, deployment configs, that run in production with elevated privileges. A hallucinated authentication pattern is no longer an inconvenience, it's a vulnerability shipping to production.

Working with the AIUC-1 Consortium and security teams at leading agentic development platforms including Lovable, Cursor, Codex, and Claude Code, we identified 75 coding-agent-specific risks across 13 thematic categories. They cluster into seven priority domains:

Secure defaults for code generation
Secrets management
Runtime execution and sandbox integrity
Dependency and supply chain integrity
Agent autonomy and human oversight
Data confidentiality and IP protection
Transparency & governance around shared responsibilities

For each area, the whitepaper outlines each risk in more detail and list the AIUC-1 requirements that are now rolled out to all coding agent platforms pursuing AIUC-1 Certification.

First coding agent AIUC-1 certifications underway

Lovable is set to become the first agentic development platform certified against the extended standard, with a third-party audit scheduled for summer 2026 conducted by Schellman. Other coding agent platforms have already begun certification.

In the whitepaper, Lovable demonstrates specific safeguards and controls implemented to meet AIUC-1 requirements, offering practical examples for other coding agent platforms pursuing certification.

Download whitepapper

Latest articles

Research

Looking ahead: Consortium feedback shaping the Q3-2026 update

Looking ahead: Consortium feedback shaping the Q3-2026 update

Over 200 peer-review comments were exchanged in the Q2 2026 update of AIUC-1. Many were implemented in the update, while other areas required more work and were carried forward as priority areas for the Q3 update.

Research

From Consortium input to new controls: AIUC-1 quarterly update process

From Consortium input to new controls: AIUC-1 quarterly update process

AIUC-1's quarterly cadence is deliberately faster than traditional standard-setting processes, enabled by a structured, repeatable process grounded in contributions from AIUC-1 consortium members, technical peer-review, and transparent change management. This quarter, over 120 Consortium members contributed through a series of technical sessions and 200+ peer-review comments.

Research

Q2-2026 update: Strengthening MCP security, agent permissions & third-party risk

Q2-2026 update: Strengthening MCP security, agent permissions & third-party risk

AI capabilities are reshaping cybersecurity faster than most frameworks can adapt - Mythos Preview is the latest proof point. AIUC-1 is updated quarterly to keep pace with shifts in technology, risk, and regulation. This quarter's refresh focused on MCP security, third-party risk management, and agent identity and permissions.