Anthropic files for IPO as AI investment and research expand

Anthropic has filed for an IPO with the SEC, seeking to go public ahead of OpenAI. Its filing discloses rapid revenue growth, widening losses, strong enterprise adoption and an unusual governance structure. It also highlights competition for electricity, bringing data center dependence on natural gas and associated climate risks into focus. Separately, SpaceX and Apollo reached a GPU financing agreement to support SpaceX’s expanding AI ambitions without relying solely on traditional debt markets.

OpenAI announced research results involving the Kakeya conjecture, the exponential time hypothesis and quantum chromodynamics, alongside mechanisms for academic researchers to validate and extend the work. In software development, Greptile is expanding defect detection beyond code review into workflow testing, while CodeRabbit is using OpenAI models to identify security risks in complex codebases and help developers assess proposed fixes.

Agent reliability and safety remain active research priorities. New studies address communication failures in multi-agent systems, learning from mistakes, unsafe consequences of underspecified requests and whether scientific agents retain model safety constraints. PaperFormalizer involves researchers in translating mathematical papers into Lean code to address gaps in proofs and assumptions. Meanwhile, The RealReal has strengthened bot defenses against automated shopping agents, and Google introduced an AI Quest education project focused on mapping the human brain.