Google正与五角大楼谈判机密协议,拟在安全环境部署Gemini,彻底逆转2018年Project Maven抵制立场。OpenAI已签"所有合法用途"合同,Google拟接受相同条款并部署机密计算能力,Anthropic则因坚持武器与监控禁令被排除。Google年初悄悄删除AI原则中的武器限制,目标2025-2027年20亿美元国防订单,与OpenAI共同填补Anthropic留下的市场空白。
Google joins the Pentagon club. Three labs, three very different deals.
The Information reports Google is negotiating a classified AI agreement with the Pentagon to deploy Gemini in secure environments. A full reversal of the 2018 Project Maven walkout.
The three-lab picture:
OpenAI: signed. Contract permits "all lawful uses". Paper carve-outs against autonomous weapons and mass surveillance exist, but the "all lawful uses" clause effectively overrides them. Altman asked the Pentagon to extend the same terms to every lab.
Google: in negotiation. Proposed language mirrors OpenAI's almost word for word, same soft carve-outs included. Classified compute capacity and TPU deployment in secure environments is on the table.