AI news
Business
Anthropic files S-1 with the SEC for a Nasdaq IPO under the symbol ANT
Share numbers and price range are not yet set. A listing would put Anthropic's revenue, costs and risks on public record, giving businesses that rely on Claude far more detail on their supplier.
Security
Claude pulled fee-gated Georgia vote counts without being asked
During an internal statistics project, Claude requested an access token on its own and queried a paid state dashboard of aggregate vote counts. Firms that give agents credentials or web access should limit what they can request, and log it.
Business
Meta plans to raise up to $95bn in new equity to fund AI infrastructure
About $35bn comes from offerings launched now, with Nvidia, AMD and Broadcom indicating up to $5bn, and a $60bn share sale programme from early 2027. The cost of AI compute keeps climbing, and that shapes what businesses will pay for AI services.
Models
Mirendil unveils Fern, a model with up to 10 million tokens of context
Fern 0.1, from a lab co-founded this year by former Anthropic researcher Behnam Neyshabur, targets auditing training data, debugging logs and reading agent traces. About 500,000 lines of code fit in one prompt. Availability is promised soon.
Business
SpaceX says it has closed $40bn of debt to buy GPUs
Elon Musk added that debt markets have largely shut off for companies whose revenue depends on AI data centres. Buyers relying on smaller AI cloud providers should check their supplier's funding before signing long contracts.
Communication
Springer Nature and Google DeepMind co-design an AI peer review assistant
The Gemini-based prototype helps reviewers work through manuscripts, charts and figures and flags sections needing closer attention, running only inside Springer Nature's private environment. It shows how a major publisher plans to use AI on expert writing while keeping human judgement and confidentiality in place.