# AI News Flash — Daily Brief **Date:** Thursday, July 9, 2026 **Type:** daily **Source:** https://ainewsflash.co/brief/53 **Editors:** Justin Bunnell (https://www.linkedin.com/in/justinbunnell/), Laz Manrique (https://www.linkedin.com/in/laz-m-5a218b81/) **Publisher:** AI News Flash (https://ainewsflash.co) **License:** Republish with attribution. --- ## Key takeaways - OpenAI ships GPT-Live, a real-time multimodal model, alongside GPT-5.6 Sol rollout. - xAI releases Grok 4.5 publicly, built with Cursor coding data after SpaceX acquisition. - Claude Cowork expands to web and mobile, adding Microsoft 365 writing tools for Max users. - Anthropic unveils Project Glasswing, deploys Mythos2 Preview on zero-days - Grok 4.5 Launches Publicly With 1.5T-Parameter V9 Foundation and Cursor Training Data - DreamDojo trains robot world model on 44k hours of human video --- ## Platforms ### OpenAI ships GPT-Live, a real-time multimodal model, alongside GPT-5.6 Sol rollout. OpenAI released GPT-Live on July 8, publishing both a product announcement and a system card on the same day. The new offering is a real-time multimodal product, adding to an already active release week for the company. Separately, the GPT-5.6 Sol global rollout is also underway, having cleared a US government review period before opening to Plus subscribers this week. The simultaneous releases signal an accelerating product cadence at OpenAI, with implications for developers, enterprise customers, and competitors who must now respond to two distinct capability launches within the same window. > Why it matters: Developers and enterprises must now evaluate two distinct OpenAI capability tiers released in the same week. - Source: https://openai.com/news/product-releases/ ### xAI releases Grok 4.5 publicly, built with Cursor coding data after SpaceX acquisition. xAI announced Grok 4.5 on July 8 and began rolling it out publicly on July 9. The model was built in conjunction with Cursor, the coding IDE whose parent company Anysphere was acquired by SpaceX in June, with that training data intended to sharpen performance on long-horizon coding and agentic tasks. xAI characterizes Grok 4.5 as its most capable model for coding and knowledge work, and Elon Musk described it as Opus-class but faster, more token-efficient, and lower cost than competing offerings. No system card, benchmark table, or independent evaluation has been published, leaving performance claims unverified at launch. > Why it matters: Developers evaluating coding models now have a new xAI option whose real-world performance remains unverified by independent benchmarks. - Source: https://x.ai/company ### Claude Cowork expands to web and mobile, adding Microsoft 365 writing tools for Max users. Anthropic has expanded Claude Cowork beyond its original desktop-only footprint, bringing the product to web and mobile platforms starting with Max plan subscribers. Remote sessions now persist even when a user's laptop is closed, making the product more practical for mobile-first workflows. The update also introduces Microsoft 365 write tools directly inside Cowork, covering drafting and sending email, managing calendar events, and creating or updating files in OneDrive and SharePoint. The combination of cross-platform availability and deep Microsoft 365 integration positions Cowork as a more direct competitor to enterprise productivity copilots from Microsoft and Google. > Why it matters: Enterprise teams on Max plans can now use Claude as a persistent, cross-platform productivity layer within Microsoft 365 workflows. - Source: https://releasebot.io/updates/anthropic/claude ### Anthropic unveils Project Glasswing, deploys Mythos2 Preview on zero-days Anthropic unveiled Project Glasswing, a defensive cybersecurity initiative built on an as-yet-unreleased model called Claude Mythos2 Preview. The model has already surfaced thousands of high-severity vulnerabilities spanning every major operating system and browser, according to Anthropic. Enterprise partners including Cisco and AWS are evaluating the model within live production security operations, and Anthropic plans to extend access to open-source maintainers so they can proactively patch critical codebases before vulnerabilities are exploited. The initiative marks a significant step toward deploying frontier AI models specifically in security research roles, with potential consequences for how the industry approaches vulnerability discovery at scale. > Why it matters: Security teams at enterprises and open-source projects may gain AI-assisted vulnerability detection before adversaries can exploit critical flaws. - Source: https://www.anthropic.com/glasswing ## Capabilities ### Grok 4.5 Launches Publicly With 1.5T-Parameter V9 Foundation and Cursor Training Data xAI made Grok 4.5 publicly available on July 9, targeting SuperGrok Heavy and Premium+ subscribers. The model is built on a 1.5-trillion-parameter V9 foundation and incorporates training data from Cursor, the coding IDE developed by Anysphere, which SpaceX acquired. xAI positions the Cursor data as a targeted enhancement for long-horizon coding tasks and claims the model delivers performance comparable to Claude Opus 4.8 at lower cost and higher token efficiency. No system card or benchmark table has been released, meaning all performance comparisons to competing models remain unverified by independent evaluations at this stage. > Why it matters: Enterprises and developers comparing frontier coding models cannot yet validate xAI's Opus-class performance claims without independent benchmark data. - Source: https://aireleasetracker.com/latest ## Technology & Research ### DreamDojo trains robot world model on 44k hours of human video DreamDojo, accepted at ICML 2026, addresses the robot-data scarcity problem by pre-training a generalist world model on 44,000 hours of egocentric human video, a dataset 15 times longer and covering 96 times more skills than previous robot-specific datasets. The model treats human video as a proxy for shared underlying physics, then fine-tunes on small-scale robot data to transfer that knowledge to physical systems. After distillation, it runs at 10.81 frames per second and supports live teleoperation and policy evaluation without requiring physical hardware. The approach generalizes to unseen objects and out-of-distribution environments, offering a practical path around the bottleneck of collecting large-scale robot demonstrations. > Why it matters: Robotics researchers can now pre-train capable world models using abundant human video instead of scarce, expensive robot demonstration data. - Source: https://arxiv.org/abs/2602.06949 ## Regulation & Policy ### Illinois governor signs Artificial Intelligence Safety Measures Act into law Illinois became the third US state to enact frontier AI safety legislation when Gov. JB Pritzker signed SB 315 on July 6, following California and New York. The law applies to AI developers with more than $500 million in annual revenue and requires them to publish catastrophic-risk frameworks, report serious incidents to state authorities within 72 hours, and submit to annual independent third-party safety audits, a requirement with no precedent at the state level in the United States. OpenAI and Anthropic both supported the legislation, which cleared the Illinois House by unanimous vote. The law intensifies pressure on Congress to advance the Great American AI Act, which includes a federal preemption clause that could supersede state-level mandates. > Why it matters: Large AI developers now face binding state-level audit and incident-reporting obligations that could set a template for federal regulation. - Source: https://news.wttw.com/2026/07/06/pritzker-signs-landmark-ai-regulation-bill-aims-mitigate-risks ### Inaugural UN Global Dialogue on AI Governance closes in Geneva with action mandate The first session of the UN Global Dialogue on AI Governance, mandated by General Assembly resolution A/RES/79/325 and co-chaired by El Salvador and Estonia, convened in Geneva from July 6 to 7, marking the first time every UN member state has gathered specifically on AI governance. Secretary-General Guterres identified four priorities: common safety standards, human-rights red lines, capacity-building for developing nations, and an AI Environmental Transparency Initiative. The accompanying Independent Scientific Panel cautioned that science cannot guarantee frontier AI will not cause catastrophic harm. The dialogue carries no binding legal authority, but its closing call for concrete national commitments before a second session in New York in May 2027 establishes the diplomatic baseline for any future treaty or international standards negotiations. > Why it matters: The Geneva dialogue sets the diplomatic foundation that could shape binding international AI safety standards in future treaty negotiations. - Source: https://www.un.org/global-dialogue-ai-governance/en ## AI Stocks ### Apple commits more than $30 billion to Broadcom for custom AI chips through 2031. Broadcom filed an SEC disclosure confirming a multi-year custom chip agreement with Apple, and on July 8 Apple placed a public figure on the deal: more than $30 billion through 2031, the largest single commitment under Apple's American Manufacturing Program. The agreement covers custom ASIC silicon intended for multiple generations of iPhone, iPad, and Mac products, and includes a $1.5 billion expansion of Broadcom's Fort Collins, Colorado manufacturing facility. Broadcom shares surged roughly 5% on the announcement, breaking above their 100-day and 200-day moving averages after a 25% pullback from a 52-week high of $495. With Apple representing approximately 20% of Broadcom's annual revenue, the long-term contract removes what analysts had identified as the stock's most significant uncertainty. > Why it matters: Locking Apple's chip spend through 2031 removes the largest revenue uncertainty for Broadcom investors and signals Apple's long-term custom silicon strategy. - Source: https://www.indmoney.com/blog/us-stocks/apple-broadcom-30-billion-chip-deal-explained ### AMD finalizes a 6-gigawatt MI450 deal with Meta ahead of its late-July AI event. AMD has finalized a deal with Meta to supply up to six gigawatts of AI computing capacity using MI450 family systems, with deployment targeted for the second half of 2026. The agreement positions AMD as the primary Nvidia alternative for large-scale data center AI workloads. AMD's data-center revenue grew more than 50% year-over-year in Q1 2026, and management has guided for approximately 70% growth in Q2. Attention is now turning to AMD's late-July Advancing AI event, where analysts at Citi and Goldman Sachs anticipate additional hyperscaler partnership announcements. AMD shares are up 153% year-to-date but trade near the Street's average price target of $518, leaving limited consensus upside unless new customer names are disclosed. > Why it matters: The Meta MI450 deal cements AMD as a credible at-scale Nvidia alternative, directly affecting data center procurement decisions across the industry. - Source: https://stockstoday.com/between-a-sector-rout-and-a-6-gigawatt-deal-amds-ai-narrative-hangs-in-the-balance/ --- Archive: https://ainewsflash.co/archive RSS: https://ainewsflash.co/rss.xml LLM index: https://ainewsflash.co/llms.txt