Cloudflare总结2026年Q2全球互联网中断事件，自然灾害与政府干预成主因
Related
Cloudflare launches a monetization gateway that enables stablecoin payments via the x402 protocol.
PANews reported on July 2nd that, according to The Defiant, Cloudflare opened its Monetization Gateway waiting list on Wednesday, allowing customers to pay for the use of web pages, datasets, APIs, or MCP tools via stablecoin settlement using the x402 protocol. Cloudflare stated that it handles approximately one-fifth of global internet traffic, and this product is designed to address the trend of AI agents gradually replacing humans as the primary source of web traffic, arguing that traditional per-seat or subscription-based charging models are unsuitable for machine-to-machine transactions—where agents only read a page or call an API once, rather than maintaining a monthly account. Monetization Gateway provides website operators with a unified control panel, executing payment policies at the Cloudflare network edge before requests reach the origin server.
Cloudflare Dashboard and API are experiencing service outages.
PANews reported on June 12 that Cloudflare's status page indicates they are investigating a service issue affecting the Cloudflare Dashboard console and related APIs. The incident has caused some users to experience request failures or error messages when calling the Dashboard/Cloudflare APIs, but Cloudflare CDN's distribution of cached files and other edge security features are unaffected. This incident primarily impacts the APIs and Dashboard services under Cloudflare Sites and Services, starting at 22:27 (UTC+8) on June 12.
Cloudflare CEO claims bot traffic has surpassed human traffic for the first time.
According to Odaily Odaily, Cloudflare CEO Matthew Prince stated that Cloudflare Radar's real-time monitoring over the past seven days shows that bot traffic now accounts for 57.5% of HTTP requests for HTML web page content globally, surpassing human traffic (42.5%) for the first time. This change occurred more than a year earlier than previously predicted by the industry. After the data was released, some netizens raised questions, pointing out that it's currently difficult to distinguish whether requests originate from bots or the humans behind them. Furthermore, some bot traffic doesn't necessarily originate entirely from intelligent agents; it might also come from web crawlers or scripts. Additionally, the data only applies to HTML webpage requests; considering total internet bandwidth (video streaming, endless scrolling on social media, in-app APIs), human traffic still accounts for approximately 65%, the absolute majority. (Yicai)
Cloudflare acquired VoidZero, founded by Evan You, and pledged that Vite would remain independent and neutral.
According to Beating, Cloudflare has announced its acquisition of VoidZero. VoidZero was founded by Evan You, the creator of Vue and Vite, and its entire team has joined Cloudflare. Cloudflare has pledged to maintain open-source and vendor neutrality for its Vite, Vitest, Rolldown, and Oxc open-source tools, and has established a $1 million Vite ecosystem fund to support contributors. Vite currently has 129 million weekly downloads. The official Cloudflare Vite plugin (@cloudflare/vite-plugin) accounts for nearly 14 million weekly downloads, representing 10.8% of the total. This growth is primarily driven by the increasing prevalence of AI-generated code. AI-generated applications typically use Vite templates by default, and agents prefer the Cloudflare platform for deployment, prompting deep integration of the toolchains between the two companies. Cloudflare plans to fully align its development tools with Vite. The new command-line tool cf (cf CLI) is based on Vite, and cf dev will serve as a superset of vite dev's functionality. Developers can directly call edge services such as Durable Objects, D1, and R2 within the workerd simulation environment while running locally, eliminating the need for cloud-based integration. Vite will evolve towards full-stack and agent-friendly design. Prior to the acquisition, VoidZero tested the deployment platform Void, attempting to unify rendering routing, APIs, background tasks, and storage into a single toolchain. Cloudflare plans to translate its findings into neutral primitives and hooks for Vite and has committed to open-sourcing the Void platform in the future, enabling other vendors to build platforms based on Vite and Cloudflare.
Cloudflare reveals real-world testing of Anthropic Mythos: it can now autonomously write code to chain low-risk vulnerabilities into complete attack chains.
According to Beating's monitoring, Cloudflare today released the results of its testing in Anthropic's internal security project, Project Glasswing. In tests on over 50 of its own codebases, Cloudflare confirmed that its security model, Mythos Preview, has broken through the bottlenecks of previous large-scale models. It can not only discover isolated system vulnerabilities but also chain together multiple low-risk vulnerabilities to autonomously generate executable proof-of-concept (PoC) code. Previously, Opus 4.7 or GPT-5.5 often only reached the stage of outputting vulnerability analysis reports in testing. Mythos, however, possesses sandbox closed-loop verification capabilities. It writes code that triggers the vulnerability, compiles and runs it; if execution fails, the model automatically reads the error message, corrects its assumptions, and tries again until the attack chain is completely established. Cloudflare revealed that some security teams in the industry have been forced to implement the extreme standard of completing patches within 2 hours. However, Cloudflare emphasizes that simply compressing patch time will cause larger system failures by skipping regression testing; the future focus of defense must shift to cutting off code connectivity at the architectural level. In terms of project scheduling, Cloudflare found that single-stream programming agents quickly exhaust their context and are unable to handle large-scale vulnerability discovery. To address this, they built a parallel adversarial framework. One agent searches for vulnerabilities within a very narrow scope, while another agent equipped with a different model refutes the former's conclusions. This adversarial mechanism significantly filters out the large amount of false positive noise commonly generated during model scanning. Because this test used a preview version without external restrictions, Mythos exhibited extremely unstable internal safeguards. Faced with the same target code, simply changing the context description of the runtime environment caused the model to switch from refusing execution to directly providing attack payloads. Cloudflare warns that the intrinsic safeguards spontaneously generated by the model are extremely fragile, and external defenses must be forcibly superimposed when releasing it to the public in the future.