AI
Nvidia bans Anthropic model over data fears
Major firms are restricting use of Anthropic's Fable after it retained logs, while others demand zero-data guarantees amid privacy concerns.
YESTERDAY, STILL WORTH KNOWING
Still
Crunchy.
AI labs face a trust crisis and a slowdown, while regulators and hardware makers race to keep up with the technology's rapid growth.
AI Labs Face a Trust Crisis
Major AI firms are facing a backlash from enterprise clients who are demanding concrete guarantees about data privacy. Nvidia, an investor in Anthropic, has restricted the use of Anthropic's Fable model to open-source projects, while Palantir has blocked its deployment until Anthropic grants irrevocable zero-data-retention guarantees. This pushback signals that customers are no longer willing to accept vague promises and are instead seeking strict control over their data.
The tension between safety and speed is becoming a defining feature of the industry. OpenAI, Anthropic and Google DeepMind have been in talks for weeks regarding AI safety, while leaders from these companies have called for external governance and a slowdown in frontier AI development. This follows warnings from an Anthropic whistleblower that the technology could pose an existential risk by the end of the decade. However, the Trump administration has reportedly dismissed these safety concerns in favour of keeping pace with China, creating a complex landscape for regulation.
Regulators Target Children and Deepfakes
The European Union is moving to tighten controls on online services for young people. A leaked draft of the EU Kids Act would require age verification for social media, generative AI and online games, aiming to protect children from addictive design and deepfakes. The proposal also includes strict parental controls, potentially setting a new standard for digital identity management across the bloc.
Law enforcement is also taking action against the commercial side of AI-generated harm. The Manhattan District Attorney has seized the domains of 12 websites hosting AI-generated non-consensual pornography, affecting more than 1,200 people. This operation targets the businesses that profit from such content and could set a precedent for future enforcement actions against the infrastructure of AI misuse.
Hardware and Energy Challenges
The physical infrastructure required to power AI is facing significant strain. US AI data centres are projected to consume 15 billion cubic feet of natural gas daily by 2035, a 117% increase, as the industry struggles to secure power infrastructure. This surge would rank US data centres as the world's fifth-largest natural gas consumer, highlighting a critical bottleneck for AI development.
Hardware manufacturers are also making significant strides in chip production. Intel has reportedly achieved an 80% yield on its 18A manufacturing node for its Panther Lake system-on-chip, a strong result that could support its foundry business. Meanwhile, MediaTek has unveiled the Dimensity 9600 Pro, claiming it is the world's first 2 nanometre smartphone processor. NVIDIA is also pushing a new metric for AI infrastructure, focusing on tokens per megawatt to improve energy efficiency.