Are You Open?

A Stanford HAI article examines arguments from Stanford HAI Denning Director James Landay that open-weight AI models, while useful, fall well short of genuine open source AI, and that the current US policy debate is missing that distinction.

The piece notes recent progress from Chinese open-weight models, including Moonshot AI's Kimi K3 and Alibaba's Qwen3.8-Max, which have narrowed the performance gap with top American models. It also references a corporate open letter from Nvidia, Microsoft, Meta and other companies warning against premature US restrictions on open-weight models, and notes the current US administration's shift toward greater scrutiny of frontier AI releases, including a Commerce Department order that took Anthropic's Mythos 5 and Fable 5 offline.

According to the article, Landay argues that releasing model weights alone, without training data, code, and tooling, amounts to "open distribution" rather than open source, since outsiders still cannot verify how a model was built or why it behaves as it does. He points to the Linux Foundation's Model Openness Framework "Open Science" tier as the bar genuine open source AI should meet.

The article also covers Landay's concerns about closed AI models, including data lock-in, concentrated blind spots around security failures, and the risk of frontier capability becoming a geopolitical bargaining chip. Landay argues universities, not companies, are best placed to build truly open frontier systems.

Mobile Identity vs SMS OTP: 5 APIs could get you there

When 15-20% of SMS one-time passwords fail to deliver, you’re not just losing security—you’re losing customers. Companies switching to network APIs report 4-5% growth uplift. SIM swap attacks are up 1,000% in some markets. SMS pumping fraud costs tens of thousands monthly. Authentication delays cause measurable abandonment. Yet enterprises remain stuck on infrastructure everyone agrees is obsolete. Bahadir “Bob” Yavuz, Head of Products at GTC, and Stewart Tinson, Project Director at AI-360 You’ll learn: • Why SMS delivery rates run 15-20% below submission rates—and what that costs you • How network APIs like Number Verify deliver verification in 1-2 seconds vs 15-20 for SMS • The Indonesia model: 200+ million users, 20% improvement, all three telcos collaborating • Which APIs can achieve both an enhanced quality of security, and quality of service to the end user Key topics: Synthetic identity detection • Number Verify vs SMS OTP • SIM swap attack patterns • GSMA Camara and OpenGateway standards • Data quality in fraud detection • GDPR privacy-by-design architecture • Telco collaboration requirements • Real-world deployment case studies • False positive rate management • Network API productization For CISOs, CIOs, CFOs, and security teams evaluating authentication strategies, this session provides the business case, technical reality, and deployment roadmap for moving beyond SMS.

BrightTALK