Microsoft and OpenAI announced an updated partnership agreement on April 30, ending their previous exclusive relationship. Under the new agreement, OpenAI can now deploy its models on other cloud platforms such as AWS, while Microsoft remains the "primary cloud partner" and will get priority access to the latest models. Microsoft will continue to receive a 20% share of all OpenAI platform revenue (subject to a cap), but no longer has to pay OpenAI 20% of Azure OpenAI service revenues. Both parties have removed the key AGI triggering clause, extending the non-exclusive licensing arrangement through 2032, with Microsoft retaining approximately 27% equity stake. This marks a shift from deep integration to a looser, financially oriented collaboration, as Microsoft strengthens its in-house AI model development and evaluates external models such as Anthropic and Gemini to reduce reliance on OpenAI.

OpenAI announced it has signed contracts securing 10 gigawatts of AI computing capacity, far exceeding its original 2029 target. This milestone significantly accelerates OpenAI's data center expansion plans, providing robust support for its growing AI model training and inference demands. The achievement aligns with OpenAI's recent strategic shift—reportedly abandoning its Stargate self-built data center project in favor of leased compute capacity. The 10GW-level compute reserve will reinforce OpenAI's leadership in generative AI and enable the development and deployment of even larger-scale models in the future.

According to Counterpoint Research, Anthropic captured a 31.4% share of the LLM market revenue in Q1 2026, surpassing OpenAI's 29%, despite having only 134 million monthly active users compared to OpenAI's approximately 900 million. Anthropic's average monthly revenue per user reached $16.20, significantly higher than OpenAI's $2.20 and Microsoft's $5.00, reflecting its commercial success in high-end professional markets. Global monthly LLM users now exceed 3.8 billion, generating quarterly revenue of about $20.7 billion. Anthropic is considering a new funding round that could value the company above $900 billion. Major tech firms are projected to invest $725 billion in AI infrastructure in 2026, a 77% increase from 2025.

The Cyberspace Administration of China has launched a four-month special campaign conducted in two phases. The first phase focuses on source governance of AI technologies, targeting seven key issues: large models operating without registration, insufficient safety review capabilities, non-compliant training data, AI data poisoning, inadequate labeling of generated content, misuse of AI for illegal activities, and lack of oversight over open-source models. The second phase targets information ecosystem governance, addressing seven types of AI-related abuses including generating "digital sludge," creating false information, impersonating others, spreading violent or vulgar content, harming minors' rights, AI-powered sockpuppet networks, and unauthorized AI products and services. A prior campaign before 2025 resulted in the removal of over 3,500 AI apps and deletion of 960,000 illegal pieces of content.

Versions 2.6.2 and 2.6.3 of the PyTorch Lightning library were found to contain Shai-Hulud, a malicious software package published directly to the PyPI repository using compromised credentials, bypassing GitHub submissions. The malicious code steals users' cloud credentials, authentication tokens, environment variables, and keys, attempting to upload them to newly created public GitHub repositories. Attackers exploited developers' trust in third-party package managers. The community recommends rolling back to the secure version 2.6.1 and emphasizes the importance of dependency pinning and sandboxed execution. This incident highlights the security risks posed by complex dependencies and weak security practices in machine learning projects.

Major Linux distributions are patching CopyFail (CVE-2026-31431), a local privilege escalation vulnerability with a CVSS score of 7.8. The flaw stems from a logic error in the kernel's authencesn cryptographic template, allowing unprivileged local users to write four controlled bytes into the page cache and gain root privileges. It can also serve as a container escape primitive affecting Kubernetes nodes. A single Python script can reliably exploit this vulnerability across mainstream distributions like Ubuntu, Amazon Linux, SUSE, and Debian—without race conditions—and is more broadly applicable than Dirty Cow or Dirty Pipe. Patches are available from Debian, Ubuntu, and SUSE, but most systems remain unpatched. The vulnerability was discovered by Theori using AI-assisted security scanning tools.

Cloudflare and Stripe have jointly introduced a new protocol enabling AI agents to autonomously create Cloudflare accounts, purchase domains, and deploy applications—all without human intervention. The protocol consists of three components: a service discovery directory, OAuth-based identity verification with automatic account creation, and tokenized billing (default monthly limit: $100). Original payment details are never exposed to the AI agent. Any platform with authenticated users can provision Cloudflare resources via a single API call. Similar to how OAuth standardized delegated access, this protocol extends standardization to payment and account creation, treating AI agents as first-class entities.

Hippocratic AI has launched Polaris 5.0, built on over 180 million real patient interactions and featuring a 5-trillion-parameter architecture (with a 700-billion-parameter core), outperforming GPT-5.4 Pro, Claude Opus 4.7, and Gemini 3.1 Pro across clinical accuracy, HIPAA compliance, and empathetic dialogue. The system delivers a 1.5-second time-to-first-word response and introduces over 20 new clinical skills, including drug safety (99.95%), clinical escalation safety (99.75%), and HIPAA certification (99.1%). It supports real-time multilingual switching and cough sound detection. Previous versions achieved 99.89% clinical guidance accuracy with zero serious harm incidents and have been validated by over 7,500 licensed U.S. physicians.

Jiliu Tech, China's largest independent AI computing cluster provider, filed for an IPO with the Hong Kong Stock Exchange on April 29. Founded in 2023 by a team from Tsinghua and Peking Universities, the company has completed 11 funding rounds totaling over 2.2 billion yuan, reaching a post-Series D valuation of 9.16 billion yuan (approximately $1.3 billion). Revenue grew from 31.8 million yuan in 2023 to 520 million yuan in 2025, representing a compound annual growth rate of 304.5%, with profitability achieved starting in 2024. As of April 2026, it manages over 90,000 GPUs with a total compute capacity exceeding 34,000 PFLOPS, and maintains a customer repurchase rate above 80%. Zhipu AI holds a 7.7% stake, making it the largest external shareholder. Proceeds will fund R&D, service expansion, and market development.

The U.S. Senate Judiciary Committee unanimously passed an AI safety bill on April 30 requiring companies like OpenAI and Meta to ban minors from using chatbots. The bill mandates that AI firms implement strict age verification systems, prohibits offering AI companion features to minors, and bans chatbots from推送 pornographic content or encouraging self-harm messages to underage users. The move responds to growing public concern about the potential harms of rapidly spreading AI technologies on youth and reflects bipartisan cooperation on online protections for children. The bill must still pass full Senate and House votes before becoming law.