Wiz 报告“黑色十月”:九个 AI 开发者平台接连披露安全事件

来源:Wiz:Research(安全研究) 2026年10月30日 06:51 AIHOT 评分:83 精选
摘要:Wiz 研究团队梳理 10 月内至少九个 AI 开发者平台披露的安全事件,包括 Fireworks AI、Lightning AI、Railway、Warp、OpenCode 等,多起报告指向凭据窃取、agent swarm 攻击和以“蒸馏”为目的的窃取。
推荐理由:Wiz 将十起平台安全事件连成一条线索,并用两个被盗密钥的滥用案例展示 LLMJacking 的变现链条。
# Wiz 报告“黑色十月”:九个 AI 开发者平台接连披露安全事件 > 原标题:“Black October” for AI Developer Platforms - 来源:Wiz:Research - 发布时间:2026-10-30T22:51:44.000Z - AIHOT:https://aihot.news/items/la0qb77ds4znf2r7ub14a3bqz - 原文:https://www.wiz.io/blog/black-october-for-ai-developer-platforms ## 摘要 Wiz 研究团队梳理 10 月内至少九个 AI 开发者平台披露的安全事件,包括 Fireworks AI、Lightning AI、Railway、Warp、OpenCode 等,多起报告指向凭据窃取、agent swarm 攻击和以“蒸馏”为目的的窃取。 ## 推荐理由 Wiz 将十起平台安全事件连成一条线索,并用两个被盗密钥的滥用案例展示 LLMJacking 的变现链条。 ## 正文 · 中文译文 本月至少有**九**家 AI 开发者平台披露了安全事件: - [Fireworks AI](https://fireworks.ai/security/bulletins/october-2026-security-incident)(10 月 6 日) - [Andromeda](https://andromeda.ai/security-updates/october-2026-security-incident)(10 月 9 日) - [Zeabur](https://zeabur.com/blogs/october-2026-security-incident)(10 月 14 日) - [Lightning AI](https://lightning.ai/blog/oct-2026-incident)(10 月 15 日) - [OpenCode](https://thehackernews.com/2026/10/opencode-console-breach-may-have.html)(10 月 16–17 日) - [Railway](https://blog.railway.com/p/security-disclosure-wednesday-october-21st)(10 月 20 日) - [Warp](https://www.bleepingcomputer.com/news/security/warp-discloses-breach-exposing-ai-agent-logs-user-telemetry-data/)(10 月 23 日) - [CommandCode](https://x.com/CommandCodeAI/status/2115583532055826770)(10 月 28 日) - [Northflank](https://status.northflank.com/)(10 月 30 日) _Wiz 客户可查阅威胁情报中心的相关公告,了解_ [_Railway 平台入侵事件_](https://app.wiz.io/boards/threat-center/wiz-adv-2026-189)_、_[_Lightning AI 安全事件_](https://app.wiz.io/boards/threat-center/wiz-adv-2026-194)_以及_[_整轮攻击活动_](https://app.wiz.io/boards/threat-center/wiz-adv-2026-199)_的更多信息。_  [Fireworks](https://x.com/lqiao/status/2108985661743325257?a=hRwgh8Url3MQ3gnyMDg&s=20)、[OpenCode](https://x.com/thdxr/status/2111790511409258796?a=kd5ip7cUye6U-c9bDCA&s=20) 和 [Warp](https://x.com/zachlloydtweets/status/2114800787716972873) 均报告了“智能体蜂群”攻击的迹象。Andromeda 的事件报告称这是“全行业攻击”的一部分。OpenCode 和 CommandCode 则明确将各自事件归因于一次“[蒸馏](https://www.bleepingcomputer.com/news/security/opencode-hack-exposes-api-keys-amid-ai-distillation-attacks/)”尝试。 换作其他任何一个月,这都会是头号安全新闻。但眼下事件接连不断,很少有报告能把这些攻击串联起来(云安全联盟是[一个值得注意的例外](https://labs.cloudsecurityalliance.org/research/csa-research-note-inference-provider-platform-chain-campaign/))。 ## 这些攻击的共同模式 目前,这些安全事件大多仍在分析中,尚无详细的技术信息。Zeabur 公布了完整的报告,[该公司将因此次事件而关停](https://zeabur.com/blogs/zeabur-is-shutting-down)。Railway 发布了[初步报告和时间线](https://blog.railway.com/p/preliminary-report-october-security-incident)。  不过,已有一些可供参考的细节。Fireworks、Zeabur 和 Lightning 都将事件的根因归结为凭据访问。Lightning [此前曾](https://lightning.ai/blog/pytorch-lightning-supply-chain-attack)受到 [Mini-Shai Hulud 攻击活动](https://www.wiz.io/blog/mini-shai-hulud-supply-chain-sap-npm)中一起供应链攻击的影响。Railway 目前表示,他们认为初始访问是通过“我们构建基础设施中的权限提升”实现的;Warp 则暗示根因是第三方供应商依赖,OpenCode 将其归因于“遗留系统”。 所有公开报告都指向针对性利用,明确在搜寻存储的环境变量,重点是那些授予 AI 访问权限的变量。尽管这些攻击之间没有单一的漏洞或依赖关系相连,但整体情况与近期两方面的报告高度吻合: [智能体攻击](https://www.wiz.io/blog/the-rise-of-agentic-attacks),能让威胁行为者更快、更高效地攻陷受害者。 [LLMJacking](https://www.wiz.io/blog/from-phishing-to-llmjacking) 的兴起,以及与之相关的 LLM 经纪商基础设施。它们形成了一个让窃取行为易于变现的市场,很可能促使更多行为者开展窃取凭据的行动。[蒸馏](https://www.anthropic.com/news/detecting-and-preventing-distillation-attacks)(尤其是针对中国 AI 实验室的指控)同样在推动需求。 ## 影响与后续活动 迄今为止,关于这些攻击影响的报告仅限于社交媒体上的零星说法,包括: - 据称利用存储在 Railway 中的加密密钥,[窃取了约 37.1 万美元](https://x.com/tayvano_/status/2113056813646238196)的加密货币 - 滥用窃取的 AI 访问权限,例如来自 OpenCode 的权限,[导致用户账单高达数百至数千美元](https://x.com/colinparsonscom/status/2111451256426250580?s=20) 不过,Wiz 已观察到**企业凭据的下游滥用**,其泄露可归因于上述某家供应商的入侵。我们将讨论一对典型的此类案例。 这两起事件中,我们都看到密钥被滥用,而这些密钥此前大量来自某家遭入侵供应商的基础设施。  两者首次未授权活动的特征完全一致,都发生在供应商泄露事件所报告的凭据被访问后**30 分钟以内**。当时只有一次来自住宅 IP 的 InvokeModel 调用,在一分钟的扫描中,用窃取的密钥测试了多家组织。该调用请求的是 Claude Fable 5.1,失败后回退到 Opus 5.5,后者成功。 在初次验证之后的几天里,两把密钥都只有零星流量:来自同一个小型 Go 客户端,每天 14 到 120 次调用。两把密钥上的调用步调一致,相隔约一分钟。这说明有一个持续运行的验证任务,在确认密钥是否仍然有效。 ### 案例 A 初次验证四天后,一个轮换代理池通过数十个地址进行了目录探测,对各家厂商的每个模型各发一次调用:Amazon Nova、DeepSeek、gpt-oss、GLM、Kimi、Grok、GPT-6 Astra 以及较旧的 Claude 模型。除 Fable 外,所有调用都成功了。几小时后,一个住宅地址上的 `boto3` 脚本枚举了 Bedrock 和 EC2 配额。22 秒后,同一地址开始出现批量流量。在接下来的半小时内,第二台主机摸清了各区域的配额和模型可用性,不到一分钟就发出了超过一千次调用。 **洪流** 在密钥被处置前的三十多个小时里,攻击者发起了约 80 万次 Bedrock 调用,其中约 90% 成功。批量流量先来自住宅地址,随后在夜间转为来自一个 Level 3 托管地址。该密钥的策略允许 `global.*` 跨区域推理配置文件,攻击者把每次调用都走了这些配置文件。由于全局路由把负载分散到各区域,只有约 6% 的调用被限流。流量严重偏向 Opus:Opus 模型占成功调用的四分之三,其中 Opus 5 占 36%、Opus 5.5 占 22%、Opus 4.8 占 10%。 LLMJacking 令牌使用情况细分 ### 案例 B 与案例 A 不同,这把密钥受到限制。初次验证七天后,它出现在一个疑似转售池中。在四个小时里,来自轮换托管地址的数千次调用失败,每个地址活跃约 15 分钟。调用失败是因为该密钥的策略只允许通过 `us.*` 跨区域推理配置文件使用 Bedrock,而该池请求的是 `global.*`。 次日清晨,一个共享 VPN 出口上的检查程序识别出了可用的 `us.*` 模型 ID。两小时后,一个 Level 3 地址上的 `boto3` 脚本重新测试了 `global.*` 和 `us.*`,并枚举了 Bedrock 和 EC2 配额。这正是探测案例 A 时所用的同一个脚本,连版本字符串都相同。几分钟内,闸门大开。 **洪流**  在密钥被处置前的十三个小时里,那一个 Level 3 地址发起了约 170 万次 Bedrock 调用,其中超过一百万次成功。其余几乎全部被限流:没有全局路由,流量撞上了账户的区域配额。流量分散在十几个 Claude 模型上,Opus 占比低于案例 A。Opus 5.5、Sonnet 4.5 和 Opus 5 合计占成功调用的四分之三。 LLMJacking 令牌使用情况细分 ### 被盗令牌 这两种情况下的 token 用量特征都与转售的编程智能体流量相似。典型调用只发送少量未缓存输入 token,这是提示词缓存的典型特征,返回的输出也只有几百个 token。而长尾调用则携带数万个未缓存输入 token,这与转售商背后有大量不同终端用户的情况相符。按标价计算,仅这两个案例**很可能就产生了超过 10 万美元的用量费用**。 ## 我们从这些攻击者身上看到的情况 我们的研究已经能够将上述案例中密钥的使用与公开的转售组织联系起来。虽然我们无法明确建立这些转售组织与最初的入侵或上文记录的“洪水”式 token 滥用之间的关系,但这些行为者确实能够访问被泄露的密钥,并从自己的基础设施对这些密钥进行了测试。多个行为者使用这些密钥,说明此类凭证存在一个活跃的市场。 ### “momo”——转售(以及入侵?) Wiz Research 识别出一个以 `23.106.46.239` 为中心的行为者,此人似乎自称“momo”,能够访问案例 A 和案例 B 中使用的密钥,并与一个公开的 LLM 转售网站有关联。与该 IP 地址关联的基础设施可追溯至 2025 年,曾托管多个标题与 LLM 转售相关的网站,其中包括目前仍在运营的 `llm-routing.com`,该网站宣传以折扣价访问前沿模型,并接受 USDT 付款。该 IP 还曾在 10 月中旬托管过一个 Artex 落地页,并被 Zeabur 公开指认为其入侵事件的一部分。虽然我们无法明确将此操作与上文记录的“洪水”式滥用联系起来,但该运营者确实能够访问被盗密钥。 - 网站 `momo.skin` 在 2026 年初解析到 `23.106.46.239`,另一个通过被动 DNS 关联的网站(`direct.465415.xyz`)使用了标题“Moの自建网盘”。  - 2026 年 3 月,该行为者开始宣传 LLM 相关服务,托管了标题为“ChatGPT Team 席位兑换”和“ChatGPT 号池管理”的网站。  - 2026 年 10 月 8 日,一个解析到 `23.106.46.239` 的子域名(`artex.nhhh.de`)托管了一个标题为“ARTEX — 自主渗透测试控制台”的页面。这比 Zeabur 时间线中确认该 IP 地址针对其基础设施的时间(10 月 12 日至 14 日)早四天。  - 10 月下旬,`23.106.46.239` 地址对案例 A 和案例 B 中使用的密钥进行了 API 密钥检查,并尝试使用案例 B 的密钥发起 PutAccountDataRetention 调用。  llm-routing.com 的截图 ### “xingjiabiapi”——第二个转售中心 Wiz 还识别出另一个利用被盗密钥的转售操作。这个操作公开存在于 `xingjiabiapi.org`,以折扣价提供对多种前沿模型的访问。与该网站关联的基础设施(`38.246.114.118`)被发现对案例 A 和案例 B 中使用的密钥进行了权限检查。两起检查都发生在来自无关 IP 地址的密钥高频滥用已经开始之后,这表明这些密钥存在一个活跃的转售市场。 - 主网站的一个子域名 `s3.xingjiabiapi.org` 从 2026 年 9 月 10 日起解析到 `38.246.114.118`。这个 `38.246.114.118` IP 地址枚举了案例 A 中使用的密钥,列出了各区域的推理配置文件和模型,并逐页查看了每一项 Bedrock 配额。它还对案例 B 中使用的密钥进行了手动测试。 xingjiabiapi.org 的截图 - 关于这个攻击者,有一点值得注意:他们似乎在试图防范蒸馏。其域名快照中托管了一个名为“蒸馏检测监控”的页面,该页面按三个类别给用户打分,并试图将他们划分为“高风险”“观察”和“正常”。这可能说明攻击者在设法避免与蒸馏扯上关系,或许是因为这类活动会导致账号被封禁。 ## 失陷指标(IOC) 这些 IP 被用于测试窃取的密钥,疑似与 LLM token 滥用和转售有关。我们目前没有证据表明它们与本文提到的“黑色十月”攻击有直接关联。 | 指标 | 用途 | 观察到的活动 | | --- | --- | --- | | 23.106.46.239 | 由 Zeabur 报告,后来被用于凭据校验 | 10/26 - 10/29 | | 38.246.114.118 | 检查了多个被盗密钥的权限 | 10/25 - 10/29 | | 104.243.32.50 | 检查了多个被盗密钥的权限 | 10/19 - 10/30 | | 31.57.201.201 | 使用多个被盗密钥进行识别 | 10/18 - 10/25 | 以下共享 IP 池也被用于滥用 Bedrock。不过,每个失陷凭据使用的 IP 地址各不相同,或者这些具体 IP 频繁轮换: | 提供商 | ASN | 观察到的活动 | | --- | --- | --- | | PEG TECH | AS54600, AS398823 | 09/10 – 10/28 | | Alibaba (US) | AS45102 | 9/10 – 10/25 | | Scloud | AS142002 | 10/03 – 10/28 | | Sharktech | AS46844 | 09/10 – 10/23 | | Level 3 | AS10753 | 09/11 - 10/29 | ## 正文 · 原文 At least **nine** AI Developer Platforms have disclosed incidents over the course of the month: - [Fireworks AI](https://fireworks.ai/security/bulletins/october-2026-security-incident) (Oct 6) - [Andromeda](https://andromeda.ai/security-updates/october-2026-security-incident) (Oct 9) - [Zeabur](https://zeabur.com/blogs/october-2026-security-incident) (Oct 14) - [Lightning AI](https://lightning.ai/blog/oct-2026-incident) (Oct 15) - [OpenCode](https://thehackernews.com/2026/10/opencode-console-breach-may-have.html) (Oct 16–17) - [Railway](https://blog.railway.com/p/security-disclosure-wednesday-october-21st) (Oct 20) - [Warp](https://www.bleepingcomputer.com/news/security/warp-discloses-breach-exposing-ai-agent-logs-user-telemetry-data/) (Oct 23) - [CommandCode](https://x.com/CommandCodeAI/status/2115583532055826770) (Oct 28) - [Northflank](https://status.northflank.com/) (Oct 30) _Wiz customers can refer to our relevant Threat Intel Center advisories for more information on the_ [_Railway Platform Breach,_](https://app.wiz.io/boards/threat-center/wiz-adv-2026-189) [_Lightning AI Security Incident_](https://app.wiz.io/boards/threat-center/wiz-adv-2026-194) _and the_ [_overall campaign_](https://app.wiz.io/boards/threat-center/wiz-adv-2026-199)_._  [Fireworks](https://x.com/lqiao/status/2108985661743325257?a=hRwgh8Url3MQ3gnyMDg&s=20), [OpenCode](https://x.com/thdxr/status/2111790511409258796?a=kd5ip7cUye6U-c9bDCA&s=20), and [Warp](https://x.com/zachlloydtweets/status/2114800787716972873) have all reported indications of an “agent swarm” attack. Andromeda’s incident report calls it part of an "industry-wide attack." OpenCode and CommandCode have explicitly attributed their incidents to an attempt at “[distillation](https://www.bleepingcomputer.com/news/security/opencode-hack-exposes-api-keys-amid-ai-distillation-attacks/).” Any other month, and this would be the top security story. However, with all the ongoing incidents, few reports are connecting the dots across these attacks (Cloud Security Alliance as a [notable exception](https://labs.cloudsecurityalliance.org/research/csa-research-note-inference-provider-platform-chain-campaign/)). ## The Pattern in these Attacks Currently, most of these security incidents are still being analyzed, and detailed technical information is not available. Zeabur has shared a full write-up, and the [company will be shutting down](https://zeabur.com/blogs/zeabur-is-shutting-down) due to this incident. Railway has released a [preliminary report and timeline](https://blog.railway.com/p/preliminary-report-october-security-incident).  However, there are some indicative details. Fireworks, Zeabur, and Lightning all attribute their incident root causes to credential access. Lightning was [previously impacted](https://lightning.ai/blog/pytorch-lightning-supply-chain-attack) by a supply chain attack within [the Mini-Shai Hulud campaign](https://www.wiz.io/blog/mini-shai-hulud-supply-chain-sap-npm). Railway so far has shared they believe the initial access was via “privilege escalation in our build infrastructure,” while Warp has suggested the root cause was a third-party vendor dependency, and OpenCode attributes it to a “legacy system.” All public reports highlight targeted exploitation, explicitly hunting stored environment variables, with a focus on those granting AI access. While there is no single vulnerability or dependency linking these attacks, the overall picture is strongly aligned with recent reports on two fronts: [Agentic Attacks](https://www.wiz.io/blog/the-rise-of-agentic-attacks), which can enable threat actors to compromise victims faster and more efficiently. The rise of [LLMJacking](https://www.wiz.io/blog/from-phishing-to-llmjacking), and the associated infrastructure of LLM Brokers. These create a market that makes the theft easily monetizable, likely driving more actors to undertake operations to steal credentials. [Distillation](https://www.anthropic.com/news/detecting-and-preventing-distillation-attacks), notably alleged against Chinese AI labs, drives demand as well. ## Impact and Follow-on activity Reports on the impact of these attacks have so far been limited to anecdotal reports on social media, which include: - Alleged theft of cryptocurrency, [to the tune of ~$371,000](https://x.com/tayvano_/status/2113056813646238196), using crypto keys stored in Railway - Abuse of stolen AI access, such as those from OpenCode, [resulting in consumer bills in the hundreds to thousands of dollars](https://x.com/colinparsonscom/status/2111451256426250580?s=20) However, Wiz has been able to observe **downstream abuse of enterprise credentials**, whose compromise we can attribute to one of these vendor breaches. We will discuss an exemplary pair of such cases. In both cases, we see abuse of keys that previously were heavily used from the infrastructure of a compromised vendor.  They shared an identical first unauthorized activity, which happened **within 30 minutes** of the reported credential access within the vendor breach. It was a single InvokeModel from a residential IP address, in a one-minute sweep that tested stolen keys across multiple organizations. The call asked for Claude Fable 5.1, failed, and fell back to Opus 5.5, which worked. In the days after this initial validation, both keys saw only a trickle of traffic: 14–120 calls a day from the same small Go client. The calls on the two keys ran in lockstep, about a minute apart. This indicates a single ongoing verification job checking that the keys remained valid. ### Case A Four days after the initial validation, a rotating proxy pool ran a catalog probe across dozens of addresses, sending one call per model across vendors: Amazon Nova, DeepSeek, gpt-oss, GLM, Kimi, Grok, GPT-6 Astra and older Claude models. Every call worked except Fable. A few hours later, a `boto3` script on a residential address enumerated Bedrock and EC2 quotas. Twenty-two seconds after that, bulk traffic started from the same address. Within the next half hour, a second host mapped quotas and model availability across regions, making more than a thousand calls in under a minute. **The Flood** Over thirty hours, after which the key was remediated, the actor made approximately 800k Bedrock calls, and about 90% of them succeeded. The bulk ran first from the residential address and then, overnight, from a Level 3 hosting address. The key's policy allowed `global.*` cross-region inference profiles, and the actor routed every call through them. Because global routing spread the load across regions, only about 6% of calls were throttled. Traffic leaned heavily on Opus: Opus models were three quarters of successful calls, led by Opus 5 (36%), Opus 5.5 (22%) and Opus 4.8 (10%). Breakdown of LLMJacking token usage ### Case B Unlike Case A, this key was restricted. Seven days after the initial validation, it surfaced in an apparent reseller pool. Over four hours, thousands of failed calls came from rotating hosting addresses, each active for about 15 minutes. The calls failed because the key's policy allowed Bedrock only through `us.*` cross-region inference profiles, and the pool requested `global.*`. Early the next morning, a checker on a shared VPN exit identified the working `us.*` model IDs. Two hours later, a `boto3` script on a Level 3 address retested `global.*` and `us.*` and enumerated Bedrock and EC2 quotas. It was the same script, down to the version string, that had probed Case A. Within minutes, the floodgates opened. **The Flood**  Over thirteen hours, after which the key was remediated, that single Level 3 address made approximately 1.7 million Bedrock calls, with over a million successful. Nearly all of the rest were throttled: without global routing, the traffic ran into the account's regional quotas. Traffic was spread across a dozen Claude models and was less Opus-heavy than Case A. Opus 5.5, Sonnet 4.5 and Opus 5 made up three quarters of successful calls. Breakdown of LLMJacking token usage ### The Stolen Tokens In both cases, the token usage profile resembles resold coding-agent traffic. A typical call sent only a handful of uncached input tokens, the signature of prompt caching, and got back a few hundred tokens of output. A long tail of calls carried tens of thousands of uncached input tokens, consistent with many different end users behind a reseller. At list price, these two cases alone **likely ran up over $100k in usage**. ## What we see from these Attackers Our research has been able to link the usage of the keys in the cases above overt reselling organizations. While we are not able to definitively establish the relationship between these reselling organizations and either the initial compromise or the “flood” token abuse documented above, the actors had access to the compromised keys and tested them from their infrastructure. The use of the keys by multiple actors suggests a robust market for this type of credential. ### “momo” - Resale (and compromise?) Wiz Research identified an actor centered around `23.106.46.239` who appears to refer to themselves as “momo” who had access to the keys used in Case A and Case B and is tied to an overt LLM resale site. Infrastructure linked to this IP address goes back to 2025 and has hosted multiple sites with titles related to LLM resale, including the site `llm-routing.com` which is currently active and advertises discounted access to frontier models, while accepting payment in USDT. This IP has also hosted an Artex landing page in mid-October and was publicly named by Zeabur as being part of their compromise. While we cannot definitively tie this operation to the “flood” of abuse documented above, the operator did have access to the stolen keys. - The site `momo.skin` resolved to `23.106.46.239` in early 2026 and another site tied via passive dns (`direct.465415.xyz`), used the title "Moの自建网盘" ("Mo's self-built cloud drive").  - In March of 2026, this actor begins advertising LLM related services, hosting sites with titles such as "ChatGPT Team 席位兑换" ("ChatGPT Team seat redemption"), and "ChatGPT 号池管理" ("ChatGPT account-pool management").  - On October 8, 2026, a subdomain (`artex.nhhh.de`) resolving to `23.106.46.239` hosted a page with the title “ARTEX — 自主渗透测试控制台" ("ARTEX — autonomous penetration testing console"). This is four days before Zeabur’s timeline identifying that IP address targeting their infrastructure (Oct 12 - 14).  - In late October the `23.106.46.239` address conducts api key checks on the keys used in cases A and B and attempts a PutAccountDataRetention call using the Case B key.  Screenshot of llm-routing.com ### “Xingjiabi” - a second resale hub Wiz also identified a second resale operation that was leveraging stolen keys. This one is publicly available at `xingjiabiapi.org` where they offer access to a wide range of frontier models for discount prices. Infrastructure tied to this site (`38.246.114.118`) was identified conducting permission checks on the keys used in both cases A and B. In both cases the checks were conducted after the high volume abuse of the key from unrelated ip addresses had already begun, suggesting that there is an active resale market for these keys. - A subdomain of the main site, `s3.xingjiabiapi.org` resolved to `38.246.114.118` beginning on September 10, 2026. This `38.246.114.118` IP address enumerated the key used in case A, listing the inference profiles and models per region, paging through every bedrock quota. It also conducted a manual test of the key being used in Case B. Screenshot of xingjiabiapi.org - One interesting note about this actor is they seem to be attempting to guard _against_ distillation. Their domain was snapshotted hosting a page titled “蒸馏检测监控” (Distillation behaviour detection) that was designed to score users across three categories and attempt to sort them into “high risk”, “watch” and “normal”. This may indicate that the attackers are attempting to avoid being associated with distillation, potentially due to that activity leading to accounts being suspended. ## IOCs These IPs were used to test stolen keys and are suspected to be linked to LLM token abuse and resale. We do not have evidence that links them directly to the “Black October” attacks mentioned in this blog. | Indicator | Use | Observed Activity | | --- | --- | --- | | 23.106.46.239 | Reported by Zeabur and later used for credential checking | 10/26 - 10/29 | | 38.246.114.118 | Checked permissions on several stolen keys | 10/25 - 10/29 | | 104.243.32.50 | Checked permissions on several stolen keys | 10/19 - 10/30 | | 31.57.201.201 | Identified using multiple stolen keys | 10/18 - 10/25 | The following shared pools were also used for Bedrock abuse. However, either each compromised credential used a different IP address or the specific IPs rotated frequently: | Provider | ASN | Observed Activity | | --- | --- | --- | | PEG TECH | AS54600, AS398823 | 09/10 – 10/28 | | Alibaba (US) | AS45102 | 9/10 – 10/25 | | Scloud | AS142002 | 10/03 – 10/28 | | Sharktech | AS46844 | 09/10 – 10/23 | | Level 3 | AS10753 | 09/11 - 10/29 |