# Global AI safety and regulation efforts

> Live situation record from CLSTR: https://clstr.news/situations/global-ai-safety-and-regulation-efforts
> Updated: 2026-09-28T21:59:19.000Z. Sources: 38. Developments: 2.

OpenAI has proposed the creation of international technical standards to manage risks from advanced artificial intelligence, specifically highlighting the dangers of recursive self-improvement (RSI). The company suggested that the United States lead a collaborative effort with AI Safety Institutes from several nations to establish unified protocols for monitoring and responding to unexpected AI behavior. OpenAI noted that while development speeds might need to be slowed to meet safety thresholds, these standards should not require government licensing or pre-release approvals.

Following these discussions, U.S. Representative Ro Khanna introduced the ‘Human Control Over AI Act’. This legislative proposal seeks to prohibit AI models from autonomously modifying their own architecture or objectives without government-mandated safeguards. Unlike OpenAI’s preference for avoiding mandatory licensing, Khanna’s bill proposes the creation of a new federal AI safety agency to oversee model training and deployment through a licensing system. The legislation includes requirements for independent auditors, sandbox testing, and emergency kill switches, while also addressing the regulation of advanced AI chips and corporate liability.

In response to legislative pressure, major developers including Google, Anthropic, OpenAI, and xAI have pledged to strengthen internal controls following a meeting with President Trump. This commitment focuses on tracking model capabilities to prevent risks in biosafety, cybersecurity, and chemical threats, supplemented by the use of independent external auditors. Meanwhile, Vice President JD Vance has cautioned against granting AI firms liability protections similar to Section 230, arguing that developers must remain responsible for their products. Vance also warned against ‘cartel-like’ behavior among leading firms, emphasizing a need to balance innovation with market competition.

## Claims

- Rep. Ro Khanna introduced legislation to ban recursive self-improving AI models until federal safety safeguards and a regulatory agency are established. (corroborated by 2 sources)
- AI companies including Google, Anthropic, OpenAI, and xAI signed an agreement to implement internal controls during model training and deployment. (single source)
- JD Vance argued that AI companies should not receive liability shields like Section 230 protections. (single source)
- The US administration seeks to prevent major AI developers from forming cartels that weaken competition. (single source)
- Khanna's proposed law includes criminal penalties for deploying AI that causes mass civilian harm or for employees who bypass safety protocols. (single source)
- Under the EU AI Act, the duty regarding AI literacy became effective on February 2, 2025. (single source)

## Timeline

### 2026-09-28: AI developers pledge enhanced safety controls amid US legislative push

Leading AI companies have pledged to enhance internal safety controls, while US lawmakers propose strict bans on recursive self-improving AI and warn against granting developers liability shields.

12 sources. https://clstr.news/cluster/ro-khanna-proposes-ai-safety-bill-to-ban-recursive-self-improvement

### 2026-09-21: OpenAI proposes global AI safety standards and international cooperation

OpenAI is calling for global AI safety standards led by the U.S., proposing unified protocols to manage risks like recursive self-improvement while emphasizing that standards should not require mandatory model-

26 sources. https://clstr.news/cluster/global-stock-markets-face-volatility-amid-derivative-expirations

---
Cite as: Global AI safety and regulation efforts. CLSTR, https://clstr.news/situations/global-ai-safety-and-regulation-efforts