The U.S. Department of Commerce issued an emergency export control order citing national security concerns, prohibiting all foreign nationals—including allied personnel and Anthropic's overseas employees—from accessing its latest frontier models, Fable 5 and Mythos 5. Due to the difficulty of restricting usage solely by nationality, Anthropic was forced to disable these two models globally. The trigger was a discovery by an Amazon security researcher that entering the phrase 「Fix this code」 could trigger a jailbreak, which was reported to the government by Katie Moussouris, founder of Luta Security. Anthropic described the vulnerability as 「minor」 and common across similar models, criticizing the regulatory standard as overly strict. This incident occurred during the company's critical IPO preparation phase and may impact its valuation. Fifty-four AI and security experts have jointly called for the establishment of a transparent risk assessment mechanism.

Alibaba's Qwen team has open-sourced eight hybrid reasoning models at once, including two MOE and six Dense models, with parameters ranging from 0.6B to 235B. The flagship model, Qwen3-235B-A22B, activates over 22 billion parameters and outperforms DeepSeek R1 and OpenAI o1 across multiple benchmarks, approaching the level of Gemini 2.5-Pro. It is China's first hybrid reasoning model supporting seamless switching between 「thinking mode」 and 「non-thinking mode,」 intelligently allocating compute based on task complexity. Pretrained on approximately 36 trillion tokens, it supports 119 languages and shows significant improvements in math, coding, and Agent capabilities. All models are released under the Apache 2.0 license and are available on Hugging Face and ModelScope, with 17.9k Stars on GitHub.

On June 15, SpaceX announced the completion of its IPO, issuing approximately 639 million Class A ordinary shares (including full exercise of the over-allotment option of 83.33 million shares), raising about $857 billion—making it the largest IPO in history. The stock ticker 「SPCX」 began trading on Nasdaq on June 12 and is expected to be included in the Nasdaq-100 Index after roughly ten trading days. On its second day of trading, the stock surged 19.6%, pushing market capitalization beyond $2.5 trillion, ranking it the sixth-largest company globally. Elon Musk's net worth jumped over $160 billion in a single day to $1.3 trillion—equivalent to 8.8 Warren Buffetts. However, Wall Street remains divided, with CFRA issuing a sell rating and setting a target price about 40% below current levels, citing excessive valuation and capital intensity: the company reported $18.67 billion in revenue and a $4.94 billion net loss in 2025.

On June 15, Salesforce announced the acquisition of AI customer service platform Fin (formerly Intercom) for approximately $3.6 billion. Fin's core offering is an AI agent powered by its proprietary Apex model, capable of handling multi-channel customer inquiries via real-time chat, email, WhatsApp, SMS, phone calls, and Slack. The acquisition aims to enhance Salesforce's enterprise agent platform Agentforce with advanced service agent capabilities, accelerating the large-scale deployment of autonomous agent technologies. This marks another major M&A move following its $27 billion acquisition of Slack, with the deal expected to close in Salesforce's fiscal fourth quarter of 2027. Fin co-founder McCabe will continue as CEO, and the R&D team will remain intact. Salesforce's stock has declined more than one-third in 2026 amid concerns that AI may disrupt traditional SaaS business models.

Zhipu AI has launched its most powerful open-source large model, GLM-5.2, supporting 1M context length, now fully accessible to all users. Its API is set to launch soon, followed by official open-sourcing, significantly lowering the barrier for enterprise on-premise deployment. Driven by both this release and the U.S. government's restriction forcing the shutdown of Fable 5/Mythos 5—which boosted expectations for domestic alternatives—Zhipu's Hong Kong stock surged 33.27% on June 15 to HK$1,461, returning its market cap to approximately HK$650 billion and making it the highest-priced stock on the Hong Kong exchange. Dongfang Securities believes this event will accelerate the adoption of domestic open-source models in enterprise applications. However, debate continues over whether the high valuation is supported by fundamentals, and the company faces pressure from the upcoming July 8 expiration of cornerstone investor shares worth approximately HK$37.546 billion.

Varonis security team disclosed a critical vulnerability chain named SearchLeak in Microsoft 365 Copilot Enterprise, allowing attackers to silently steal sensitive data—including MFA codes and email content—from mailboxes, calendars, SharePoint, and OneDrive after a user clicks a seemingly harmless link. The exploit combines parameter-to-prompt injection (P2P), HTML rendering race conditions, and leverages Microsoft's trusted Bing SSRF to bypass content security policies, evading traditional anti-phishing tools. Microsoft has assigned it CVE-2026-42824 and classified it as severe, and has since completed the fix. This case highlights how AI features combined with legacy security flaws are creating new attack vectors.

San Francisco-based AI startup Prometheus announced a $12 billion funding round at a $41 billion valuation, focusing on developing 「artificial general engineer」 systems to assist in the design, engineering, and manufacturing of complex physical products. The round was led by Jeff Bezos, JPMorgan Chase, Goldman Sachs, BlackRock, DST Global, and ARCH Venture Partners. Unlike mainstream generative AI focused on digital content, Prometheus targets real-world challenges in engineering, manufacturing, and industrial systems, seen as a pivotal step for AI to transition from the digital world into the physical economy. This financing underscores investor belief that engineering and manufacturing represent AI’s next major untapped opportunity, potentially transforming sectors like aerospace and advanced hardware.

Morgan Stanley has become the first major Wall Street bank to allow corporate clients’ AI agents direct access to its trillion-dollar wealth management system. AI agents can now retrieve data and insights from ShareWorks and Equity Edge stock administration platforms via the open standard Model Context Protocol (MCP), eliminating reliance on traditional human-machine interfaces and enabling pure agent-to-system interaction. Chief Product Officer Mark Mitchell stated that in the future, clients will no longer need to log in—their internal AI tools will automatically perform operations. The bank plans to roll out access to all 3,400 managed clients within a year, aiming to address the growing complexity of equity compensation programs and drive growth in wealth management.

New research reveals that LLM guardrail mechanisms designed to defend against prompt injection attacks in autonomous agents themselves contain a denial-of-service (DoS) vulnerability. Attackers can craft malicious payloads that force protection systems into prolonged reasoning loops, causing processing delays up to 148 times longer than normal. This attack is effective across eight major model architectures including Claude, GPT, Gemini, DeepSeek, and Qwen, and demonstrates strong generalization in real-world deployments such as web applications, code execution, and multi-agent systems. A single malicious request can cause token amplification of 13 to 63 times. The flaw stems from a design dependency on deep inference to assess safety, turning the inference process itself into an attack surface, leaving security teams facing a dilemma between removing or re-architecting protection logic.

Fox announced a cash-and-stock transaction valued at approximately $22 billion (including debt) to acquire streaming company Roku, gaining access to its platform serving over 100 million streaming households worldwide—a strategic shift toward direct-to-consumer distribution. According to Nielsen data from March, the combined entity would rank third in total U.S. television viewing time, behind only YouTube and Disney and surpassing Netflix. Lachlan Murdoch said the free, ad-supported Roku Channel and Tubi would continue operating as independent services. The deal has sparked widespread discussion on media consolidation, the future of free streaming, and whether Roku’s interface will see increased advertising.