网络安全
Binance 详情假工作面试和哈克的AI工具妥协
在 Google 上的首选来源中添加 Unite.AI
Binance 称在2026年11月3日公布的初步调查结果中认为，在2026年10月29日盗取了约63.4亿美元的数字资产之前，假招聘采访，恶意开发者软件和一个已失密的内部AI工具。
交易所将吸收财务影响，而不是传递给用户。 据Binance报告，在2026年10月31日，为弥补损失拨出了超过100%的资金，补充了98%受影响的象征性资产，冻结或拦截了约28.5亿美元，但需视调节和市场价格波动情况而定。 它正在购买未清、流动性较低的资产。
补充资金时引用的储备证明页描述了1:1对保管用户资产的支持。 它使用Merkle-tree记录和zk-SNARK证明，这样用户就可以验证其余额是否包含在负债报告中。
调查结果仍然是初步的。 调查正在进行中，独立审查待决，比南斯隐瞒一些技术细节，以避免帮助恶意行为者策划袭击。
钱包袭击前的月份
2026年6月，一个钱包DevOps工程师在试图安装一个通过搜索引擎找到的流行的AI编码工具的同时下载了一个被篡改的Homebrew安装。 宾恩斯检测到异常点，更换了设备并旋转了相关证书，但工程师的角色和钱包基础设施团队的同事已经暴露在攻击者身上。
2026年9月，另一个软件工程师通过合法的求职平台被假扮为项目主任的人联系。 与一位自称人力资源代表的现场视频访谈之后，接近尾声时，工程师奉命将包含恶意代码存储器的任务下载到工作计算机上。 它的代码在设置中执行。
恶意软件以恶意版本取代了合法的pyenv组件，并与攻击者控制的服务器通信。 Binance 称端点保护没有检测到当时的零天恶意软件，而攻击者则采取步骤隐藏活动并清除妥协的痕迹。
2026年10月20日，攻击者使用第二位工程师失密的计算机和一个内部AI工具，将控制扩展至原DevOps工程师和另一位钱包软件工程师。 随后在DevOps工程师的计算机上发现了一个集装箱部署许可标志。 协调世界时10月29日17:15至19:56，攻击者登录到第三工程师的电脑中，使用从DevOps工程师那里偷来的证书，部署一个容器，并触发了钱包攻击的发布的恶意脚本。
综合封装
攻击者利用未经授权的进入一个内部钱包签名服务，在Binance正常退出程序之外启动交易。 在协调世界时20:05至20:09之间，五个比南斯热钱包将233.46万XRP送到攻击者控制的地址。 随后的流出来自索拉纳和乐观热钱包，一个比南斯热钱包将18,270比特币送到攻击者控制的地址。
攻击者还利用了Ethereum的EIP-7702授权机制，让账户将某些执行权下放给智能合同，而不暴露其私人钥匙。 50个签名交易操纵了账户非ce，并确立了恶意授权，赋予攻击者控制的合同对Binance热钱包的部分执行权。 这些授权使资产能够继续独立于交易所的提款基础设施。
因此，暂停退出和关闭热钱包服务并没有使现有的链上许可失效。 由于钱包使用了多方计算，所以没有单一的私人密钥可供立即撤销，因此需要一个需要额外时间的专门程序。 在协调世界时23:56完成最后的未经授权的转让，即2,961.9以太，在此之前仍未完成；此后没有观察到其他未经授权的转让。 2026年10月30日协调世界时00:10,Binance从它的Binance14热钱包执行新的授权到零地址，清除了恶意代表团代码。
重建系统和联邦调查局的单独归属
截至11月3日，比南斯在新的环境下重建了受影响的钱包系统，并重建了受影响的内部AI工具及其访问控制。 它还更换了受影响的设备、退役的相关证书和信使、完成了对雇员设备的强化扫描、扩大了基于AI的行为异常检测和加强钱包服务风险控制。
正在聘请独立的第三方网络安全专家验证调查结果，并提出其他措施建议。 Binance正在增加对AI辅助检测，端点保护，开发者-环境安全和行为监测的投资。 它还在努力建立一个安全研究基金，并与AI公司、行业参与者和保安公司一道，采取有针对性的举措，重点关注AI在区块链环境中的安全。
在2026年10月31日的一份单独通知中，联邦调查局的称朝鲜贸易商行为者负有责任，该局及其USSS和IRS-CI合作伙伴观察到他们主要将被盗资产转换成以太，预计通过DeFi和即时游览服务会进一步洗钱。 这些机构敦促节点运营商、交易所、DeFi服务、区块链分析公司和其他虚拟资产供应商阻止涉及或源于所列地址的交易。
该账户留下了一个超出恶意软件检测的封存问题：更换第一个失密设备并没有结束攻击者接触钱包团队，中止退出并没有取消链上的权限。 独立审查现在成为Binance重建这一链条的检查，以及它重建的控制是否解决了这一问题。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Cybersecurity
Binance Details Fake Job Interviews and AI Tool Compromise in Hack
Add Unite.AI to your preferred sources on Google
Binance said in initial findings published November 3, 2026, that fake recruitment interviews, malicious developer software and a compromised internal AI tool preceded the theft of approximately $6.34 billion in digital assets on October 29, 2026.
The exchange will absorb the financial impact rather than pass it to users. Binance reported setting aside more than 100% of the funds required to cover the loss, replenishing 98% of impacted token assets as of October 31, 2026, and freezing or intercepting approximately $2.85 billion, subject to reconciliation and market-price volatility. It is purchasing the outstanding, less-liquid assets.
The Proof of Reserves page cited for the replenishment describes 1:1 backing for custodial user assets. It uses Merkle-tree records and zk-SNARK proofs so users can verify that their balances are included in the liabilities report.
The findings remain initial. The investigation is ongoing, independent review is pending, and Binance is withholding some technical details to avoid helping malicious actors plan attacks.
Months of Access Before the Wallet Attack
In June 2026, a wallet DevOps engineer downloaded a tampered Homebrew installation while attempting to install a popular AI coding tool found through a search engine. Binance detected the anomaly, replaced the device and rotated related credentials, but the engineer’s role and colleagues on the wallet infrastructure team had already been exposed to the attacker.
In September 2026, another software engineer was contacted through a legitimate job-hunting platform by someone posing as a project director. A live video interview with a purported human-resources representative was followed by a second interview in a popular social chat app with someone claiming to be the CEO of a fictitious company. Near its end, the engineer was instructed to download an assignment containing a malicious code repository onto a work computer. Its code executed during setup.
The malware replaced a legitimate pyenv component with a malicious version and communicated with an attacker-controlled server. Binance said endpoint protection did not detect the zero-day malware at the time, while the attacker took steps to conceal the activity and remove traces of the compromise.
On October 20, 2026, the attacker used the second engineer’s compromised computer and an internal AI tool to extend control to the original DevOps engineer and another wallet software engineer. A container-deployment permission token was then found on the DevOps engineer’s computer. Between 17:15 and 19:56 UTC on October 29, the attacker logged into the third engineer’s computer, used credentials stolen from the DevOps engineer, deployed a container and triggered the malicious script that launched the wallet attack.
On-Chain Authorizations Complicated Containment
The attacker used unauthorized access to an internal wallet-signing service to initiate transactions outside Binance’s normal withdrawal process. Between 20:05 and 20:09 UTC, five Binance hot wallets sent 233.46 million XRP to an attacker-controlled address. Subsequent outflows came from Solana and Optimism hot wallets, and a Binance hot wallet sent 18,270 Bitcoin to an attacker-controlled address.
The attacker also exploited Ethereum’s EIP-7702 authorization mechanism, which lets an account delegate certain execution rights to a smart contract without exposing its private key. Fifty signing transactions manipulated the account nonce and established malicious delegation authorizations, giving an attacker-controlled contract partial execution rights over a Binance hot wallet. Those authorizations allowed assets to keep moving independently of the exchange’s withdrawal infrastructure.
Consequently, suspending withdrawals and shutting down hot-wallet services did not invalidate the existing on-chain permissions. Because the wallet used multi-party computation, there was no single private key available for immediate revocation, requiring a specialized procedure that took additional time. The final unauthorized transfer, 2,961.9 Ether, finalized on-chain at 23:56 UTC after previously remaining pending; no further unauthorized transfers were observed afterward. At 00:10 UTC on October 30, 2026, Binance executed a new authorization from its Binance14 hot wallet to the zero address, clearing the malicious delegation code.
Rebuilt Systems and a Separate FBI Attribution
As of November 3, Binance had rebuilt the affected wallet system in a new environment and rebuilt the affected internal AI tool and its access controls. It also replaced affected devices, decommissioned related credentials and tokens, completed enhanced scans across employee devices, expanded AI-based behavioral-anomaly detection and strengthened wallet-service risk controls.
Independent third-party cybersecurity experts are being engaged to validate the findings and recommend additional measures. Binance is increasing investment in AI-enabled detection, endpoint protection, developer-environment security and behavioral monitoring. It is also working to establish a security research fund and targeted initiatives with AI companies, industry participants and security firms focused on AI security in blockchain environments.
In a separate notice dated October 31, 2026, the FBI said North Korea’s TraderTraitor actors were responsible, that the bureau and its USSS and IRS-CI partners had observed them converting stolen assets primarily into Ether, and that further laundering was expected through DeFi and instant-swap services. The agencies urged node operators, exchanges, DeFi services, blockchain analytics firms and other virtual-asset providers to block transactions involving or derived from the listed addresses.
The account leaves a containment problem that extends beyond malware detection: replacing the first compromised device did not end the attacker’s exposure to the wallet team, and suspending withdrawals did not remove on-chain authority. The independent review now becomes the check on both Binance’s reconstruction of that chain and whether its rebuilt controls address it.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.