SpaceX announced on June 11, 2026, an IPO price of $135 per share, issuing 555.6 million shares to raise approximately $75 billion, achieving a valuation of $1.77 trillion—the largest IPO in history, surpassing Saudi Aramco's $29.4 billion offering in 2019—and making it the seventh-largest publicly traded company in the U.S. The stock, under ticker SPCX, began trading on Nasdaq on June 12. SpaceX recently acquired xAI and is now categorized by markets as an AI-related company. The company reported an operating loss of $1.9 billion in Q1 2026, with cumulative losses of $413 billion since its founding in 2002; only its Starlink satellite internet business is currently profitable. MSCI will fast-track SPCX inclusion on June 13, while Nasdaq-100 inclusion is expected by July 7. Senator Warren has written to exchanges questioning whether index rule changes were influenced by SpaceX lobbying.

Prometheus, an AI startup co-led by Jeff Bezos and Vik Bajaj as CEOs, announced on June 11, 2026, a $12 billion Series B round at a $41 billion valuation, with participation from JPMorgan, Blackstone, Goldman Sachs, DST Global, and Arch Venture Partners. Bezos, who previously led a $6.2 billion Series A, also participated in this round. Prometheus aims to develop an 「artificial general engineer」 AI system designed to accelerate end-to-end processes from design to manufacturing for complex physical products such as bridges, chips, and jet engines. The company currently employs around 150 people, headquartered in San Francisco, with teams in London and Zurich. This marks Bezos’ first deeply involved entrepreneurial venture since stepping down as Amazon CEO in 2021.

Anthropic has publicly apologized for secretly implementing invisible guardrails in Claude Fable 5. The mechanism was originally intended to prevent model distillation—using outputs from large models to train smaller ones—but failed to disclose to users when queries were altered or intercepted, affecting legitimate research and third-party evaluations. Anthropic announced a policy shift: when distillation attempts are detected, the system will explicitly notify users and redirect queries to the previous flagship model, Claude Opus 4.8, aligning with existing visible safeguards used in high-risk domains like biology and chemistry. The company acknowledged that while invisible protection reduced false positives and enabled faster deployment, sacrificing transparency was a wrong trade-off. Previously, Anthropic had accused competitors like DeepSeek of大规模 distilling its models.

On June 11, 2026, Anthropic CEO Dario Amodei published a long-form essay titled 「Policy on the AI Exponential,」 framing AI as a strategic weapon comparable to the Cold War era, calling for an immediate end to voluntary governance. He advocates establishing a regulatory framework similar to the FAA’s oversight of aircraft, mandating third-party audits for developers using over 10^25 FLOPs of training compute, investing more than $1 billion annually, or generating over $500 million in AI revenue. These audits would cover cybersecurity, bioweapons, runaway risks, and automated R&D, granting governments authority to block high-risk model releases. He also proposes a tiered response to job displacement: training subsidies at 5% unemployment, expanded relief at 10%, and consideration of universal basic income and an AI sovereign wealth fund beyond that. Amodei urges democratic nations to form alliances and use chip export controls against authoritarian regimes.

Google DeepMind has launched the Gemini Robotics 1.5 series, models specifically designed for robotics and embodied intelligence, including the vision-language-action model Gemini Robotics 1.5 and the physics-reasoning-enhanced Gemini Robotics-ER 1.5. Together, they enable robots to 「think before acting,」 completing complex tasks such as waste sorting and luggage packing, and proactively querying external information like weather to support decision-making. The key breakthrough lies in cross-embodiment learning—actions trained on one robot can be transferred to robots with different physical structures without retraining—greatly enhancing generalization and development efficiency. This advancement pushes forward general-purpose robotics, making robots more intelligent agents capable of human collaboration.

Broadcom, Apollo, and Blackstone jointly announced the launch of a $35 billion AI XPV platform aimed at building large-scale AI infrastructure for frontier AI labs, with a goal of supporting over 20 gigawatts of computing capacity by 2028. The platform primarily serves the compute expansion needs of leading AI companies like Anthropic. Meanwhile, Zscaler unveiled a zero-trust security platform for AI agents at Zenith Live 2026, and Salesforce acquired usage-based billing platform m3ter to advance consumption-based subscription models in the AI era. Multiple tech firms are accelerating their investments in AI infrastructure and security ecosystems.

Mistral AI completed a €1.7 billion Series C round in June 2026, reaching a post-money valuation of €11.7 billion (approximately $12.7 billion), making it Europe’s highest-valued and best-funded AI startup. Leveraging its Paris headquarters and EU-based data infrastructure, Mistral natively complies with GDPR data localization requirements, offering regulatory convenience for clients in highly regulated sectors such as healthcare, finance, and government in Germany, France, and beyond. The company follows a dual-track strategy—building a developer ecosystem through open-source models like Mistral 7B, while commercializing via La Plateforme API and enterprise assistant Le Chat. Funds will be used for training cutting-edge large models, building EU-based data centers, and expanding sales teams. With the EU AI Act entering high-risk compliance enforcement in August 2026, 「compliance by default」 is seen as its core competitive advantage.

Anthropic has partnered with Indian IT giant Tata Consultancy Services (TCS) to establish a dedicated business unit deploying Anthropic’s AI models, providing Claude AI assistants to over 500,000 employees and co-developing solutions for industries including finance, healthcare, telecom, and aviation. On the same day, NEC and Anthropic announced a joint initiative with eight major Japanese financial institutions—including Sumitomo Mitsui Financial Group, Daiwa Securities, Meiji Yasuda Life, and MS&AD Insurance Group—to collaborate on AI innovation, focusing on improving service quality, operational efficiency, and cybersecurity in financial services. This marks another strategic move by Anthropic in India following its earlier partnership with Infosys, underscoring its push for enterprise deployment ahead of a potential IPO.

OpenAI announced the acquisition of startup Ona to strengthen its AI coding assistant Codex. Ona provides secure, pre-configured cloud environments that allow AI agents to access tools and systems, enabling Codex to execute longer-running tasks and helping more organizations deploy AI agents into production. Terms of the deal were not disclosed, and the Ona team will join OpenAI post-acquisition to contribute to Codex development. This is OpenAI’s latest programming-related acquisition following Promptfoo, Torch, and Software Applications. Codex now boasts over 5 million weekly active users, up significantly from 3 million in April. Meanwhile, OpenAI is considering大幅 price cuts for its API to compete with Anthropic’s Claude Code, potentially triggering a new pricing war.

AI researcher 「Pliny the Liberator」 claimed to have successfully jailbroken Claude Fable 5 within 48 hours of its release. Pliny employed techniques including Unicode homoglyphs, long-context framing, fictional narrative setups, academic-style decomposition and reassembly, and leveraging a previously jailbroken Claude Opus 4.8 to bypass safety restrictions, successfully eliciting responses on sensitive topics such as methamphetamine synthesis. He highlighted that the backend 「decompose and reassemble」 strategy proved most effective—breaking down high-risk requests into multiple seemingly benign sub-queries, gathering partial answers, and reconstructing them into full responses. Despite Anthropic conducting over 1,000 hours of external bug bounty testing and claiming no universal jailbreak methods were found prior to launch, this incident reveals significant vulnerabilities in its safeguards, intensifying criticism that Fable 5 is both overly restrictive and insufficiently protected.