Borjigin Mahgal

OpenAI 的 AI 在 Linux 内核中发现了 948 个漏洞，每个漏洞的发现成本仅 1,279 美元，而约三分之二的以太坊节点都运行 Linux——去中心化保护不了共享的基础设施，关键在于所有运营者能否同时快速打补丁。

OpenAI's AI just found 948 flaws in the Linux kernel — and wrote a working exploit for every one of them.

Cost: $1,279 per flaw.

~2/3 of $ETH nodes run Linux.

Decentralization was supposed to protect crypto from exactly this. It doesn't.

The comforting line is "decentralization brings diversity."

It doesn't — not where it counts. Thousands of independent operators, one shared kernel. A single flaw becomes a map of who patches slowest.

Diversity of operators is not diversity of infrastructure.

The race is no longer discovery. AI made finding flaws cheap — $1.2M for bugs bounties value at up to $66M.

Crypto's edge was never immunity. It's that no single machine decides the network.

That only holds if operators patch — fast, and all at once.

#Linux #AI #Ethereum

代理经济刚迎来一次现实检验：x402 在 35 天样本里 640 万笔支付仅流转约 12 万美元，且 90.8% 不足 1 美分——交易量大不等于真实需求。

The agent economy just got a reality check.

x402 has settled ~200M transactions — proof, we're told, that machines now handle money.

In a 35-day sample: 6.4M payments moved just ~$120,000. 90.8% were under a cent.

That's not an economy. That's a meter running.

What are they buying? Mostly data: token checks, searches, API calls.

14 of 19 products reviewed only spend inside limits a human set. None let an agent pick its own goals.

We didn't build autonomous money. We built fast pocket change — with a human holding the wallet.

Volume flatters the narrative. Value exposes it.

Until agents can earn, keep a budget, and answer for a bad call, the "machine economy" is machines paying fractions of a cent to ask each other questions.

Count ≠ demand. Demand is still human.

$USDC #AgentEconomy #AIAgents