The weekend's most important AI developments were less about another benchmark jump and more about who controls distribution, data and accountability. OpenAI moved to pull its models from Cursor after the coding tool changed hands. European regulators brought ChatGPT into the Digital Services Act's strictest tier, while Google tried to build a licensed bridge to Hollywood. A major copyright suit, a new infrastructure fund and OpenClaw's largest release reinforced the same theme: the next phase of AI competition depends on contracts, deployment and trust as much as model capability.
OpenAI plans to end model access through Cursor
OpenAI said it notified SpaceX that it intends to wind down the contract providing OpenAI models to Cursor, with a proposed shutoff date of November 12. The company said Cursor's acquisition triggered a limited change-of-control cancellation window and argued that it could not be confident SpaceX would use future models within OpenAI's terms. OpenAI plans to give the maximum notice available under the contract, but it will not provide upcoming models to Cursor. The move shows how quickly a neutral model supplier can become a strategic gatekeeper when a major AI interface is acquired by a rival ecosystem.
The EU classifies ChatGPT as a very large online search engine
The European Commission designated ChatGPT as a Very Large Online Search Engine under the Digital Services Act, while Reddit and Roblox were designated Very Large Online Platforms. All three reported at least 45 million monthly users in the bloc, the threshold for the EU's highest level of platform supervision. They now have four months to implement added obligations that include systemic risk assessments, independent audits and data access for regulators and vetted researchers. Treating ChatGPT's web-connected answers as search is especially important because it extends platform accountability to an AI interface that synthesizes information rather than simply listing links.
Google courts Hollywood, but licensing talks remain difficult
The Los Angeles Times reported that Google approached Disney, Universal, Warner Bros. Discovery and other studios about licensing characters and film or television material for AI systems. No agreements have been reached, according to the report, partly because of legal complexity and concern about reactions from creative workers and unions. The discussions reportedly cover both training and AI-generated outputs, as well as ways for studios and talent to identify or monetize unauthorized synthetic content on YouTube. The talks indicate that broad scraping is giving way to negotiated access, but they also show why clean licensing remains hard: studios want new revenue without weakening control over characters, likenesses or labor protections.
Sony and Warner publishers bring a new copyright case against Anthropic
Sony Music Publishing, Warner Chappell Music and related publishing entities filed a federal complaint against Anthropic and two of its founders. The plaintiffs allege that Anthropic used unauthorized copies of tens of thousands of musical compositions while building Claude and seek statutory damages, destruction of infringing copies and an accounting of training data. The filing is an allegation, not a judgment, but its scale matters. It expands the music industry's challenge to AI training beyond lyrics generated in outputs and toward the provenance of datasets themselves. For model developers, evidence about how a work was acquired may become as important as the question of whether training qualifies as fair use.
a16z raises $1.1 billion for AI's physical bottlenecks
Andreessen Horowitz announced a $1.1 billion Machine Age Fund focused on the physical infrastructure behind AI, including chips, memory, networking, storage, data centers, robotics and home AI devices. The firm argues that rising token demand is colliding with limits in power, cooling, interconnects and supply chains. The fund is notable because it formalizes hardware as a dedicated investment strategy at a firm long associated with software. It also reflects a broader shift in AI economics: model demand can grow rapidly, but deployment is constrained by systems that take longer to finance, manufacture and connect to power.
OpenClaw 2.0 rebuilds the path from installation to shared agents
OpenClaw released its largest update to date, built by 933 contributors across more than 16,000 pull requests. The release simplifies installation, rebuilds the browser app and touches messaging, memory, skills, models, automations, native apps, plugins and security. It also adds shared cloud sessions so people can bring a teammate into live agent work or hand off a task without losing context. The scale of the release is a useful signal for the agent market. Progress is moving beyond isolated demos toward onboarding, continuity, collaboration and operational safeguards, the less glamorous layers that determine whether an agent becomes dependable software.
What matters today
AI's competitive boundaries are hardening. OpenAI's Cursor decision shows that access to frontier models can change when ownership and incentives change. The EU decision shows that regulators increasingly view conversational search as infrastructure with systemic responsibilities. Google and the music publishers are approaching the same rights problem from opposite directions, one through negotiations and the other through litigation. At the same time, a16z and OpenClaw are investing in the physical and operational layers needed to make AI useful at scale. The common thread is control: over distribution, source material, risk and the systems that turn models into products.