开源AI编码工具OpenCode遭黑客攻击 用户账户产生未授权费用

开源AI编码工具OpenCode在周六遭到安全漏洞影响，约1.4%的OpenCode Console用户受到波及1。攻击者通过突破旧系统成功获取了用户存储的API密钥，导致用户在OpenAI、Anthropic和Google AI账户上产生未授权费用1。部分用户报告单个账户的损失达数百美元，少数账户损失甚至超过1000美元1。

OpenAI、Anthropic和Google已采取措施阻止泄露的API密钥继续被滥用，其中OpenAI为遭受攻击的用户进行了退款1。在应对此次攻击的防御过程中，OpenCode团队发现了当前AI安全防御面临的一个结构性问题：AI模型因安全过滤器限制而拒绝分析恶意代码，这阻碍了防御工作的展开1。为解决这一问题，OpenCode后来获得了OpenAI的Daybreak Blue程序访问权限，该程序提供的模型在安全防御工作中不太可能拒绝请求1。总统科学技术顾问委员会联合主席David Sacks指出，安全防御所需的模型和攻击分析面临着同样的AI防御墙问题1。

OpenCode, an open-source AI coding tool, experienced a security incident on Saturday that compromised the accounts of 1.4% of its OpenCode Console users 1. Attackers exploited a vulnerability in legacy systems to gain access to stored API keys, which they then used to generate unauthorized charges across users' OpenAI, Anthropic, and Google AI accounts 1.

The financial impact varied significantly among affected users, with some reporting individual account charges reaching hundreds of dollars, and a smaller number exceeding $1,000 1. In response, OpenAI, Anthropic, and Google invalidated the compromised API keys 1. OpenAI additionally issued refunds to users who had been targeted in the attack 1.

During the incident response, the OpenCode team uncovered a structural weakness in current AI security defenses 1. AI models refused to analyze malicious code during defense efforts because the code triggered built-in safety filters 1. To address this limitation, OpenCode subsequently gained access to OpenAI's Daybreak Blue program, which provides models less likely to refuse requests during security defense work 1. David Sacks, co-chair of the President's Council of Advisors on Science and Technology, noted that models needed for defensive security work face the same AI safety barriers as those used in attack analysis 1.