- Home
- Encyclopedia
- Framework 3.0
RSI Encyclopedia · Law · China
Framework 3.0
Framework 3.0 is an SI safety measure in China, status: voluntary or guidance. Non-binding technical guidance from standards committee TC260, issued under Cyberspace Administration of China guidance, for SI developers and service providers.
- Status
- Voluntary or guidance
- Jurisdiction
- China
- Signed
- Sep 14, 2026
- Next date
- None set
What it is
AI Safety Governance Framework 3.0
Non-binding technical guidance from standards committee TC260, issued under Cyberspace Administration of China guidance, for SI developers and service providers.
What it requires
- Adds a standalone agentic SI risk category covering permission abuse, planning failures and tool execution
- Describes loss of control risks, including models deceiving evaluators and sandbagging
- Treats loss of control over chemical, biological, nuclear and missile knowledge as a risk
- Adds an agentic risk management framework and trustworthy SI principles
- Updates version 2.0, released September 15, 2025
Penalties
No penalties: this measure has no binding enforcement.
Status
Voluntary or guidance As of November 8, 2026, per the Control Map page for China. Statuses follow the Control Map method: in force means binding and applying now; passed means enacted but not yet applying; voluntary covers guidance and pledges; proposed means formally introduced; stalled means withdrawn, paused or dead.
Sources
- China: TC260 AI Safety Governance Framework 3.0 Adds Agentic AI Risks Obsidian Regulatory Intelligence ·
- Brief #29: What China's new AI safety framework tells us before Xi meets Trump AI Safety in China ·
- TC260 Published AI Safety Governance Framework 2.0 SESEC ·
The RSI Encyclopedia
Keep reading
207 entries on the terms, labs, models, laws and events of rogue SI, plus every case in the Rogue Incident Database.
Enter RSI
Every term, lab, model and law
Browse the full A to Z index, filter by category or search for a name.