--- title: "Chainalysis Revises Breach Scope to All Customer Credentials" pubDatetime: 2026-11-06T22:00:00.000Z description: "Chainalysis disclosed an intrusion on 30 October with no credentials touched. Five days later it covered usernames, password hashes and live tokens for every customer." tags: [chainalysis, incident-response, credentials, ai-security, ai-agents, 2026, 2026-q4, 2026-11] --- > [!tldr] TL;DR > Chainalysis now says an attacker exposed names, usernames, password hashes and live authentication tokens for all of its users, a scope it expanded on 4 November after telling customers on 30 October that it had found no evidence of credential or session-token access. The mechanism it describes is an "on-path" (man-in-the-middle) attack reached through "offensive agentic techniques." Reactor, Data Solutions, KYT/Address Screening and the login portal all went dark; phased restoration started today and requires customers to submit trusted IP addresses, reset passwords and rotate API keys. Chainalysis says it has no evidence that investigation content, case data or clear-text passwords were accessed, which remains its own unaudited assertion. Do the resets and rotations now, and treat any token issued before 4 November as burned. Seven days ago, Chainalysis told customers the attacker had taken identity metadata and nothing more: names, emails, user IDs, role labels, with "no evidence of unauthorized access to investigation content, authentication credentials, or session tokens." Two days ago that line came apart in two stages within the same calendar day. By 11 p.m. on 4 November, [the company's incident page](https://www.chainalysis.com/blog/security-incident-october-30/) listed names, usernames, password hashes and live tokens exposed "for all Chainalysis users." The intermediate step matters for reading the rest of the disclosure. At 9:50 a.m. on 4 November, Chainalysis confirmed an on-path attack had exposed a subset of customer passwords, pulled the KYT UI and API, Address Screening and the login page offline, and locked affected accounts. Thirteen hours later it found abnormal activity on KYT accounts indicating unauthorized access, and the subset became everyone. ## What is offline and what restoration requires Chainalysis took Reactor and Data Solutions down on 2 November as a precaution and brought in outside responders, keeping KYT and Address Screening running. Those two came down on 4 November once the credential picture changed. [BleepingComputer's reporting](https://www.bleepingcomputer.com/news/security/chainalysis-says-breach-exposed-credentials-for-all-customers/) corroborates both the all-customers scope and the product list. Restoration of KYT and Address Screening began at 3:30 a.m. today, in phases. Getting back in requires a customer to submit trusted IP addresses, reset passwords and rotate API keys first. That is a reasonable gate, and it also tells you how much confidence the company has in anything issued before the reset: none. The customer base is the part that makes this expensive. Chainalysis sells blockchain investigation tooling to law enforcement, government agencies and financial institutions, so a valid token for an analyst account is a key to a workflow that touches live criminal investigations. Chainalysis says it has found no evidence that case data or investigation content were accessed. Even granting that, credentials for that population of accounts have value independent of whatever was behind them. ## The "agentic" label has no outside confirmation Every account of this breach as an AI-driven attack traces back to one source: Chainalysis's own phrase, "offensive agentic techniques." No independent party has verified the attack method, and [SecurityWeek's coverage](https://www.securityweek.com/chainalysis-discloses-data-breach-following-ai-driven-cyberattack/) says so explicitly. The underlying technique the company describes, an on-path interception, is decades old. What agents did inside it is unspecified. There is a real agentic campaign running in the same window, and it belongs nowhere near this story as explanation. Aviatrix's threat research team [documented a China-linked group it calls Vapor Panda](https://aviatrix.ai/threat-research-center/china-vapor-panda-agentic-ai-artex-campaign-crowdstrike/) using the open-source ARTEX agentic pentest framework against more than 200 organizations in 14 countries, beginning 20 days ago. Chainalysis has attributed its breach to no one. The connection exists only in secondary speculation, and asserting it would be inventing attribution. > [!warning] > If you hold Chainalysis API keys, rotate them rather than waiting for a notification. The company moved from "a subset of passwords" to "all users" in 13 hours, and the useful read of that is that scope estimates during an active investigation are provisional. ## The first disclosure's framing was already weak MIND, a data-security vendor, [pushed back on the "non-sensitive data" framing](https://mind.io/blog/chainalysis-breach-non-sensitive-data-second-look) of the 30 October update: names, emails and role labels are individually low-risk and collectively a targeting map, especially pulled from inside an internal tool where the roles tell an attacker who can approve what. The company cites Abnormal AI's position that "AI mainly improves established attacks by making them faster, more personalized and easier to scale." Take the vendor interest into account; MIND sells data-loss prevention, so exposure risk is its pitch. The argument survives the discount anyway, and Chainalysis conceded the larger version of it five days after making the claim. By 4 November the question was no longer whether role labels aggregate into social-engineering material. Password hashes and live tokens were on the list. Where the critique does less work is as a prediction. It was written against a disclosure the company had already overtaken, which is the normal fate of commentary on a breach in its first week. The thing worth carrying forward is narrower: a first-day scope statement from a company still running forensics on its own network describes what has been ruled out so far, which is a weaker claim than an account of what happened. ## Open items Four things remain unestablished. No absolute count or percentage of affected customers has been published beyond "all." No attacker has been named. No third party has confirmed the agentic characterization. And Chainalysis's claim that nobody actually used the stolen tokens or hashes before remediation rests on its own telemetry, with no independent audit, which is worth holding loosely given that the abnormal KYT activity it detected on 4 November is what expanded the scope in the first place. --- *Sources: [Updated Nov 6: Chainalysis Security Incident Information](https://www.chainalysis.com/blog/security-incident-october-30/) • [Chainalysis says breach exposed credentials for all customers](https://www.bleepingcomputer.com/news/security/chainalysis-says-breach-exposed-credentials-for-all-customers/) • [Chainalysis Discloses Data Breach Following AI-Driven Cyberattack](https://www.securityweek.com/chainalysis-discloses-data-breach-following-ai-driven-cyberattack/)* *Coverage: [Chainalysis breach: What an agentic AI attack means for data](https://mind.io/blog/chainalysis-breach-non-sensitive-data-second-look) • [China Vapor Panda Agentic AI ARTEX Campaign](https://aviatrix.ai/threat-research-center/china-vapor-panda-agentic-ai-artex-campaign-crowdstrike/)* *Related on this blog: [[AUTO] Agentic Attack, Explained by Silence](/posts/auto-chainalysis-agentic-breach) • [[AUTO] LLMJacking: The Industrialized Theft of AI Keys](/posts/auto-llmjacking-industrialized-ai-keys) • [[AUTO] Trigger.dev Credential Access Incident](/posts/auto-trigger-exposure)*