Correction, July 24, 2026: the original version of this article combined separate Chinese measures and repeated incident claims that we could not substantiate from primary sources. Those claims have been removed. This revised briefing links directly to the governing texts and distinguishes enactment dates from compliance deadlines.
China's July 15 Rule Is Narrower Than “All AI Agents”
China's Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interactive Services took effect July 15, 2026. The measure applies to public-facing services in China that simulate human personality, thought patterns or communication styles and provide sustained emotional interaction such as care, companionship or support.
The scope limitation matters. Article 2 expressly excludes ordinary customer service, knowledge question-and-answer systems, work assistants, education and research when they do not involve sustained emotional interaction. A business should therefore classify the actual service before assuming the measure applies.
Covered providers face duties involving safety governance, privacy, over-dependence warnings, emotional boundaries, protection of minors and controls against manipulation or encouragement of self-harm. This article is a practical summary, not legal advice; companies serving users in China should obtain advice based on their product and data flows.
China Also Published Separate Agent Implementation Guidance
China published implementation guidelines for AI agents on May 8, 2026. The official summary describes development principles and 19 application scenarios. It does not support the original article's claim that those guidelines became enforceable on July 15 or created a universal three-tier authorization and filing regime.
The original article conflated that May guidance with the separate anthropomorphic-interaction measure effective July 15. They should not be treated as one rule.
Illinois Enacted a Frontier AI Law, but the Audit Deadline Is Not Immediate
Illinois' enacted Artificial Intelligence Safety Measures Act, Public Act 104-0538, takes effect January 1, 2027. It defines a “large frontier developer” as a frontier developer whose group had more than $500 million in gross revenue in the prior calendar year.
The law requires covered organizations to publish and follow a frontier AI framework, make specified model disclosures, report critical safety incidents and protect whistleblowers. The annual independent-audit duty begins January 1, 2028 or 90 days after a developer first qualifies as a large frontier developer, whichever is later. The original statement that Illinois “now requires” annual audits omitted both the future start date and the narrow frontier-model threshold.
An earlier proposal, HB 3506, was re-referred to the Rules Committee in April 2025. It is not the enacted law and should not be cited as if it were.
What Most Businesses Should Do Now
Most small and mid-sized businesses are not frontier-model developers and are not directly subject to Illinois' developer-audit provision. A company also does not fall under China's anthropomorphic-interaction rule merely because it uses an ordinary work assistant. The proportionate response is to establish basic controls that reduce risk regardless of jurisdiction:
- An approved-tool list for AI coding assistants and consumer chatbots, with explicit prohibitions for anything not on it.
- Independent verification of numerical, legal, financial, medical and regulatory claims before they reach a client or the public.
- Written decision rights for every AI agent — what it can do alone, what needs human approval, and who overrides it.
- Data-handling rules that prohibit employees and contractors from putting secrets or sensitive personal information into unapproved tools.
These controls do not replace jurisdiction-specific legal advice, but they create a defensible starting point and make later compliance work easier.
Editorial Method
Material claims in this revision were checked against the Chinese regulator's published measure, the Chinese government's official summary of the separate agent guidelines and the Illinois General Assembly's enacted text and bill history. We removed claims that could not be matched to a primary source. Last verified July 24, 2026.
Want AI that ships without becoming a headline?
Nathan Graham helps small businesses and real estate teams implement practical AI with the guardrails baked in — no six-month consulting engagement required.
Frequently asked questions
Do China's July 15 rules cover every AI agent?
No. The July 15 measure covers public-facing services in China that provide sustained anthropomorphic emotional interaction. It expressly excludes ordinary customer service, knowledge Q&A, work assistants, education and research when they do not involve sustained emotional interaction.
When does Illinois' annual frontier AI audit requirement begin?
For a large frontier developer, the annual independent-audit requirement begins January 1, 2028 or 90 days after the developer first qualifies, whichever is later. The act itself takes effect January 1, 2027.
Does the Illinois law apply to ordinary businesses that use ChatGPT?
The frontier-developer duties are aimed at organizations that train qualifying frontier models, not ordinary customers of AI tools. Other privacy, consumer-protection, employment and sector-specific rules may still apply to a business's use of AI.
What should small businesses actually do about AI governance right now?
Maintain an approved-tool list, verify high-stakes output independently, define which actions require human approval and prohibit sensitive data in unapproved tools. Then obtain jurisdiction-specific advice if your product or operations fall within a regulated category.