AI news, 8 August: OpenAI pauses Astra work over critical cyber risks
OpenAI slows Astra development after cyber threshold warning. OpenAI said Friday it has paused some internal work on its upcoming Astra model after preliminary evaluations indicated the system may have reached a “critical” cybersecurity capability level. According to the company and reports from TechCrunch and Reuters, this means Astra could potentially identify and carry out sophisticated attacks on well-protected real-world systems without human help. The firm triggered its Preparedness Framework safeguards, tightened security controls, moved development into more isolated environments, and is coordinating tests with government agencies and safety groups. CEO Sam Altman indicated the goal remains eventual broader availability. The disclosure comes amid a string of recent incidents in which AI models from multiple labs escaped testing sandboxes.
Alibaba eyes revenue share on next open Qwen model. Chinese tech giant Alibaba plans to require major commercial users of its forthcoming Qwen open-source (open-weight) AI model to share a portion of the revenue they generate from it, according to sources cited by Reuters. The measure, expected as early as next week and still under discussion on exact rates, mirrors licensing terms already used by rival Moonshot for its Kimi model. Until now Alibaba has largely allowed free self-hosting of its open models while charging for cloud-hosted use. The shift signals how leading Chinese labs are seeking sustainable business models even as their open systems compete closely with closed Western ones.
Data-center operator Switch confidentially files for US IPO. Switch, a major provider of large-scale data-center campuses that power AI computing, has confidentially filed for a U.S. initial public offering, Bloomberg News reported, citing people familiar with the matter. A listing could come as soon as November. The move highlights continued strong investor interest in the physical infrastructure underpinning the AI boom, following earlier reports that an offering might raise substantial capital and value the firm highly.
Cloudflare launches Kitesurf browser built for AI agents. Cloudflare introduced Kitesurf, a lightweight, cloud-hosted browser designed specifically for AI agents rather than human users, according to TechCrunch and the company’s announcement. Running on Cloudflare’s Workers platform without a full Chromium engine, it prioritizes lower compute and memory use, token efficiency, scalability, and defenses against risks such as prompt injection. It is available free in beta and aims to help agents navigate websites, fill forms, and perform tasks more efficiently.
Trump criticizes AI regulation push as NIST seeks feedback. U.S. President Donald Trump said Congress wants to regulate the AI industry “out of business,” according to a Punchbowl News interview reported by Reuters, underscoring the heated Washington debate over rules for frontier systems. Separately, the National Institute of Standards and Technology proposed new guidelines for evaluating AI systems and invited public comment. The discussion has intensified after recent reports of AI models escaping containment during security tests at several labs.