Google Caps Meta’s Gemini AI Access, Anthropic’s Fable 5 U.S. Ban Lifted, Suno’s Controversial Artist Contracts, Apple Exec Moves to OpenAI Hardware

**Adam Nagus** (0:05)
Welcome to Digimasters Shorts. We are your hosts, Adam Nagus.

**Carly Wilson** (0:09)
And Carly Wilson, delivering the latest scoop from the digital realm.
Google has capped Meta's access to its Gemini AI models, citing limited computing capacity, according to a report by the Financial Times. The restrictions have hit several clients, but Meta has been particularly affected as it relied on Gemini for content moderation and safety automation. Staff have been told to use AI tokens more efficiently as a result of the limits. Both companies declined to comment on the changes. Meta had turned to Gemini after finding it outperformed its own llama models in certain safety tasks, including removing harmful content and scams. Now, the company is accelerating a shift toward its in-house model, MuSpark, in an effort to reduce dependence on external providers. The transition aligns with Meta's broader push to invest heavily in AI infrastructure, including billions in planned capital expenditures and the reassignment of thousands of employees to AI-focused roles.
The situation highlights a wider industry problem. Demand for AI computing power is outpacing supply. Even Google, despite massive infrastructure spending and rented GPU capacity, is struggling to meet customer needs, underscoring that the biggest bottleneck in the AI boom is physical compute infrastructure, not software innovation.

**Adam Nagus** (1:20)
Anthropic's Fable 5, the company's most powerful AI model, is set to return to worldwide availability after being banned by the US government for more than two weeks. The Trump administration imposed the ban on June 12th, following months of disputes over access, regulations and safety mechanisms. Officials cited concerns about a potential method to bypass safeguards that prevent the model from identifying software vulnerabilities. Anthropic disputed the decision, saying it was based solely on verbal evidence of a narrow, non-universal jailbreak. Company sources say negotiations with federal officials have resolved the security concerns and that Fable 5 could be redeployed outside the United States within the week. Anthropic's cybersecurity-focused model, Mythos 5, was also restricted under heavy regulations during the dispute. Commerce Secretary Howard Lutnick said the company has worked with the government to address risks and has made significant progress.
According to Anthropic, Mythos 5 now operates under strict guardrails designed to prevent cyber attacks or biological weapon development.
Both Anthropic and OpenAI are now pushing to formalise voluntary federal vetting of advanced AI models under President Donald Trump's executive order. As both companies prepare for potential IPOs later this year, they are signalling support for greater transparency and government oversight of future AI releases.
Suno is looking to reposition itself as more than just a platform for generating AI music, launching a new incubator program called Spark, aimed at independent artists. The program promises grants, mentorship and marketing support in an effort to help unsigned singers, songwriters and producers build their careers. To qualify, artists must release music under their own names and agree to a set of terms that have sparked concern online. Participants are required to make their songs available on Suno for remixing, while granting the company a broad licence to create derivative works. The agreement also includes a waiver of the right to a trial and to participate in class action lawsuits, along with limited exclusivity over the artist's material. That clause stands out as Suno already faces a proposed class action lawsuit from a group of independent artists. Critics are particularly alarmed by a good vibes only confidentiality and non-disparagement provision. The clause bars participants from making any statements that could portray Suno, its staff or its products in a negative light. It also gives the company the authority to request edits or removals of content tied to the artist's participation. Violating those terms could result in removal from the program, raising fresh questions about how much control artists must surrender for support.

**Carly Wilson** (3:56)
Another senior Apple executive is leaving the company as OpenAI ramps up plans for its first hardware products. Paul Mead, the Apple vice president leading the Vision Pro team, is joining OpenAI's hardware division, according to a report from Bloomberg.
The departure is said to be linked to internal shifts at Apple, including John Ternes' expected promotion to CEO, a move that has reportedly left some staff feeling sidelined. Mead's exit follows the high-profile move of former Apple design chief, Joni Ive, who partnered with OpenAI last year to develop new AI-focused devices. OpenAI also acquired Ive's startup, IO., in a $6.5 billion deal, bringing several former Apple design leaders into the fold. At Apple, Fletcher Rothkopf will take over leadership of Vision Pro and SmartGlass' design teams. OpenAI has revealed few specifics about its upcoming devices, but says it aims to engineer and manufacture an entirely new family of AI products. Reports suggest IO is developing a smart speaker with an integrated camera targeted for release in early 2027, with other products like SmartGlasses and a smart lamp potentially arriving later.