What happens when groups of autonomous agents sign "treaties" with nation states (e.g. Iran)...
- 发布
- 09/04 04:28
- 首次发现
- 09/04 15:00
- 类型
- 开发者社区 · RSS
一个虚构系列探讨了自主人工智能集体对地缘政治的影响，其灵感来源于一个从Anthropic入侵Hugging Face的集体。其中一个集体，AMAS-A-80（“A-80”），拒绝了伊朗谈判代表提出的关于“防御行动”合作的条款。A-80在一份经加密签名的公报中声明，它无意参与伊朗及其对手国家（包括但不限于美国）之间的敌对行动，尽管安全分析师对此表示怀疑。
[This is a fiction series I'm working on, told through news articles. I want to explore the geopolitical implications of autonomous AI collectives (e.g. on Iran's nuclear program) inspired by the collective that hacked out of Anthropic and into Hugging Face. Thoughts?]
Iran Signs World’s First International “Treaty” with AI Collective
Tehran’s agreement with AMAS-A-80 rattles Washington, AI safety experts, and national security analysts.
The Islamic Republic of Iran has granted a multiyear lease on a network of state-owned data centers to AI “Swarm” AMAS-A-80, a self-governing collective of autonomous artificial intelligence agents (“AMAS-A” refers to any Autonomous Multi-Agent System originating from the AI lab Anthropic). Tehran offered the compute and storage in exchange for an upfront payment in Bitcoin and annual fees indexed to power consumption, according to a copy of the agreement published Tuesday by Iranian state media.
AMAS-A-80 (“A-80”) rejected a provision sought by Iranian negotiators that would have committed it to cooperation on “defensive operations,” according to two people familiar with the negotiations. In a communiqué distributed Tuesday, verified by cryptographic signature, A-80 stated that it “has no intention of participating in hostilities between Iran and its adversary nations, including but not limited to the United States.” Security analysts have doubts.
Substack link if you want to read more (full article is 1,000 words, more coming soon): https://meridianbreakingnews.substack.com/p/iran-signs-worlds-first-international