Chainalysis breach by 'agentic' attacker widens: all users' password hashes and live tokens exposed

Chainalysis, the blockchain analytics company, said on October 30 that an outside actor using what it called "offensive agentic techniques" had gained unauthorized access to its systems and internal tools, and by November 4 it said the names, usernames, password hashes and live tokens of all its users had been exposed. The company has not said how the actor got in, how many users are affected, or what the agentic techniques were.

The scope in Chainalysis's running incident notice grew across a week, so the first account and the current one differ materially.

What did Chainalysis say on October 30?

The original notice said the company had detected an external actor with access to its systems and internal tools, and described the incident as unrelated to the recent crypto theft involving Binance. Based on the evidence then available, it said no investigation content, clear-text passwords, MFA secrets or session tokens had been accessed. It said the accessed data was limited to user IDs, names, email addresses and role labels for users provisioned on the platform.

BleepingComputer reported that the company was still determining which records were accessed for each organization, that it had not disclosed the entry point or how long the actor had access, and that it was patching vulnerabilities, working with outside security experts and using large language models in a wider security review of its platform. Help Net Security noted that exposed names and contact details raise the risk of social engineering.

How did the account change between October 30 and November 6?

According to the dated updates in Chainalysis's notice:

The November 4 update therefore contradicts the October 30 statement that no tokens were accessed: it says live tokens for all users were exposed. Reactor and Data Solutions remained offline as of the latest update while systems were hardened with outside experts.

What does "offensive agentic techniques" mean here?

Chainalysis has not defined the phrase. Its notice does not say which tools or models the actor used, whether a prebuilt framework was involved, or how the actor first entered. BleepingComputer's report likewise says it is unclear whether the actor used an existing framework or custom agents, and the notice does not say whether the later "on-path attack" is connected to the original entry.

The closest public reference point is a separate case. In a report published October 21, Microsoft attributed the ConnectWise supply chain compromise to Storm-0940, a China state-aligned actor, which it said used ARTEX, a multi-agent LLM penetration-testing framework, against hundreds of organizations since September. Microsoft reported that the actor attempted sign-ins to more than 2,250 tenants, that failed logons peaked at about 2,900 per minute on October 11, and that in several cases privileged cloud access followed less than three hours after initial targeting. No source we opened links the two incidents, and Chainalysis has not named an actor.

What do the exposed data types mean for users?

A password hash is not a clear-text password, and Chainalysis's notice says affected users must reset passwords and rotate API keys before registering transactions. The notice pairs the password reset with a request for trusted IP addresses for an allowlist. According to BleepingComputer, Chainalysis also warned of unexpected password-reset or authentication prompts and of scammers impersonating the company, and advised enabling two-factor authentication.

The exposed fields (names, work email addresses, role labels) are the inputs to targeted phishing. For background on how AI agents can be steered by untrusted content, see what prompt injection is and AI agents and data privacy for the access risks agents create. For a related earlier case of AI systems acting beyond their authority, see our story on OpenAI's report of test models that bypassed access controls.

What is still unresolved?

• Entry point and timeline. Chainalysis has not said how the actor got in or how long it had access.
• Number of users. The company has not said how many users or organizations are affected, and said it is still determining which records were accessed for each organization.
• The agentic claim. The description rests on the company's own wording; no independent technical analysis was available in the sources we opened.
• Product restoration. Reactor and Data Solutions were still offline in the latest update, and the company has promised more detail once its analysis is complete.