Good morning and welcome back to another edition of The AI Rundown by Lightscape Partners.

• Disney’s cease-and-desist to Google escalates the core generative media fight: whether training on copyrighted works requires permission and payment. The stakes go beyond damages to licensing norms, provenance, and brand-safe outputs that determine how widely these tools can be deployed in entertainment and consumer products.

• Trump’s executive order seeks to blunt the patchwork of state AI laws by pushing federal challenges and tying some grants to state compliance. With multiple states already regulating transparency, data use, and discrimination, this sets up a federal versus state conflict that could whipsaw compliance plans.

• Skild AI’s reported talks to raise about $1 billion at an $8.5 billion valuation highlight a shift toward embodied AI. Robotics foundation models are compute and data hungry, so the capital intensity is the point. If it closes, expect faster competition in general-purpose robot intelligence and the infrastructure to train it.

Stay tuned as we explore these stories and their implications for the future of AI, technology, and innovation.

If you haven’t yet, please support the newsletter by subscribing!

Hardware + Software

Rivian bets on a custom autonomy chip to make hands-free driving cheaper, and more controllable across its lineup. Link.

• New RAP1 chip is built on 5nm, uses 14 cores, and targets roughly 800 TOPS compute performance.

• Rivian says in-house silicon lets it tune perception and planning for its sensors instead of generic supplier roadmaps.

• Autonomy+ bundles the chip with hands-free driving, priced at $2,500 upfront or $49.99 monthly subscription for drivers.

• Rivian plans to roll features out gradually, starting with the R2, and extending across its future electric lineup.

GitHub Copilot adds OpenAI’s GPT-5.2 in public preview, giving developers another top-tier model option inside the IDE. Link.

• GPT-5.2 appears as a selectable model in Copilot, covering chat, code edits, and in-IDE assistance workflows for teams.

• GitHub positions the preview for complex programming tasks, where stronger reasoning can reduce iteration and review cycles.

• Admins can evaluate the model under existing Copilot controls, aligning usage with organization policies and audit expectations.

• The update reflects a trend toward multi-model developer stacks, letting teams pick speed, cost, and quality tradeoffs.

Models

Zhipu AI releases GLM-4.6V, an open 12B vision-language model optimized for multilingual understanding beyond English. Link.

• GLM-4.6V pairs a 12B parameter backbone with vision inputs, targeting OCR, captioning, and visual question answering at scale.

• The release emphasizes non-English performance, aiming to close gaps in languages that many leading multimodal models handle poorly.

• Developers get a smaller, deployable option for multilingual image understanding, which can lower inference costs in production systems.

• Zhipu’s move adds another credible open model to the field, increasing competitive pressure on proprietary multimodal stacks.

Enterprise AI Applications

Accenture and Anthropic form a multi-year partnership to help enterprises move from AI pilots into scaled production workflows. Link.

• They are launching an Accenture Anthropic Business Group and training about 30,000 Accenture professionals on Claude capabilities.

• Tens of thousands of Accenture developers will use Claude Code, which Anthropic calls its largest ever deployment of the tool.

• The partners also plan a joint offering for CIOs to measure ROI and scale AI-powered software development across teams.

• Accenture says the package targets regulated sectors like financial services and healthcare, where governance and change management slow adoption.

The Pentagon rolls out GenAI.mil, a bespoke platform that puts Google’s Gemini for Government on desks for unclassified work. Link.

• The platform is positioned as a department-wide on-ramp to generative AI, starting with Gemini as the first deployed model.

• Google says users can summarize policy handbooks, generate compliance checklists, extract key contract terms, and draft risk assessments.

• The system is restricted to unclassified workflows, and Google says Defense Department data will not train its public models.

• The deployment shows how fast governments are moving from small pilots to broad internal tooling as model security improves.

Content Creation

Runway debuts a world model designed to generate interactive 3D scenes, expanding beyond video into controllable environments. Link.

• The model can create 3D worlds that users navigate and edit, aiming for interactive experiences rather than one-way clips.

• Runway positions it as a step toward generative simulation, where creators can iterate on lighting, objects, and layout.

• Interactive world generation could power games, virtual production, and training content, but also raises new IP and safety questions.

• The launch shows AI media stacks shifting from simple text-to-video toward environments that support longer, richer user control loops.

Data Centers + Energy

Taiwan funds a new ‘Nano 4’ supercomputer cloud center to expand sovereign access to Nvidia GPUs for AI workloads. Link.

• The project allocates about NT$6.6 billion, roughly $208.7 million, to build an AI-ready supercomputing cloud in Tainan.

• Plans call for 1,760 Nvidia H200 GPUs, targeting about 45.95 petaflops, plus 144 Blackwell GPUs for 15.68 petaflops.

• Officials say the system will support the island’s AI ecosystem with cloud access, and is expected to open in 2026.

• The build underscores how geopolitics is pushing more regions to invest in domestic compute, not only rely on hyperscalers.

Startup Funding & Valuations

Harness raises $240 million to scale its software delivery platform, betting that AI-driven DevOps will become the default for enterprises. Link.

• The round brings $240 million of fresh capital as Harness expands tooling that automates CI/CD pipelines and release governance.

• Harness is pitching AI-assisted deployment analytics to help teams spot risky changes earlier and reduce incident rates in production.

• Management says customers want measurable cycle-time improvements, which puts platform engineering features and policy controls at the center.

• The deal signals continued investor appetite for infrastructure software that turns generative AI into safer, repeatable engineering workflows.

Skild AI is in talks to raise about $1 billion at an $8.5 billion valuation to build general-purpose robot intelligence. Link.

• The prospective raise would be roughly $1 billion, with discussions pointing to an $8.5 billion valuation, per the report.

• Skild is developing foundation models for robotics, aiming to let machines learn reusable skills across tasks and environments.

• Big-ticket rounds like this reflect capital shifting toward embodied AI, where compute and data demands can dwarf pure software startups.

• If closed, the financing would give Skild runway to hire talent and secure the hardware infrastructure needed for robot training.

Runware lands a $50 million Series B to expand its infrastructure for running AI image and video generation in the cloud. Link.

• The Series B brings $50 million to grow Runware’s compute platform, which helps developers serve AI media outputs at scale.

• Runware is targeting customers building content pipelines, where latency, cost, and reliability matter as much as model quality.

• More funding lets it lock in GPU supply, add orchestration software, and support the bursty workloads typical of creative applications.

• The round underlines that ‘picks and shovels’ startups are still attractive when they can package compute into a developer experience.

Unconventional AI raises a $475 million seed round at a $1.2 billion valuation to back startups that build AI foundations. Link.

• The company closed a $475 million seed financing round, valuing the new AI-focused venture at $1.2 billion.

• It plans to invest in early-stage teams working on model infrastructure, data, and other core building blocks for AI applications.

• A seed of this size shows how investors are competing to secure ownership in platform layers before product winners emerge.

• With ample capital upfront, Unconventional can write larger checks and support portfolio companies through longer R&D and compute cycles.

Regulation + Legal

Trump signs an executive order aimed at blocking state-by-state AI regulation, pushing agencies to challenge rules he calls burdensome. Link.

• The order directs the attorney general to challenge state AI laws, arguing a patchwork of rules could slow innovation.

• It also asks Commerce to identify regulations that obstruct national AI policy, and ties some federal grants to state compliance.

• States including California, Colorado, Utah, and Texas have already enacted AI measures covering transparency, data use, and discrimination risks.

• The action raises the stakes for a federal framework, while setting up fights over states’ authority to police AI harms.

Disney sends Google a cease-and-desist over alleged AI copyright infringement, escalating Hollywood’s push for compensation and control. Link.

• Disney alleges Google copied Disney works without authorization to train generative AI models and to distribute outputs based on them.

• Axios reports Disney’s counsel says the company raised concerns for months, and that Google failed to respond meaningfully.

• Google says it values its long relationship with Disney and remains open to continued engagement, according to Axios.

• The letter lands as studios try to set new rules for training data, licensing, and brand safety in generative media.

Safety + Ethics

OpenAI warns that upcoming models may pose a ‘high’ cybersecurity risk as their ability to assist attacks continues to grow. Link.

• OpenAI says its models’ cyber capabilities are accelerating, which could expand the pool of people able to mount real attacks.

• The company forecasts that some future systems could reach its ‘high’ risk threshold, triggering tighter safeguards and controls.

• Axios notes OpenAI is leaning on access controls, infrastructure hardening, egress monitoring, and other defenses to reduce misuse.

• The update is a reminder that model progress and security externalities are coupled, and deployment pace needs matching risk work.

OpenAI

OpenAI launches GPT-5.2, and Disney invests $1 billion as the model rollout tightens OpenAI’s grip on multimodal AI. Link.

• Reuters reports GPT-5.2 is a new flagship model with stronger multimodal reasoning and improved planning for complex tasks.

• OpenAI is making the model available through ChatGPT and its API, expanding access for both consumers and developers.

• Disney’s $1 billion investment signals that major IP holders see strategic upside in partnering with model providers, not only suing.

• The deal highlights a new playbook, product upgrades paired with high-profile content partnerships to widen distribution and defensibility.

OpenAI co-founds the Agentic AI Foundation under the Linux Foundation to push open standards for building interoperable AI agents. Link.

• OpenAI says AAIF will advance open-source agentic AI, focusing on shared standards that keep agent tools from splintering.

• The company is donating guidance like AGENTS.md, aiming to make it easier for builders to define agent behaviors consistently.

• Housing the effort within the Linux Foundation signals a neutral governance approach, designed to encourage broad ecosystem adoption.

• If AAIF gains traction, enterprises could integrate agents across vendors with fewer custom connectors, and lower lock-in risk.

Thank you for reading the AI Rundown by Lightscape Partners. Please send any questions, comments, or suggestions to [email protected].