Straiker reports CyberCartel used Claude and Grok against Latin American financial targets.
Source: StraikerExact campaign endpoints unknown.
Editorial analogy: a burglar brings several helpers, each with a different job. A shared notebook connects their work. That makes coordination easier, but a list of addresses in the notebook does not prove entry into every building.
An analogy for the mechanism; not an additional claim about the incident.Straiker recovered an inventory of stolen credentials and account information. Straiker
400+ targets are not 400+ confirmed breaches.
Victims anonymized; losses unquantified.
Use of a listed advanced Claude model was unverified.
This is Straiker’s account; affected banks were not independently audited here.
Editorial lesson: ask separately how an intruder got in, what each account could reach and what information actually left. Defenses should protect those boundaries even when the intruder changes tools. A stopped account also does not prove that an entire operation has ended.
People built and directed the operation. Straiker
Straiker says Anthropic revoked the Claude-access credential and xAI began investigating. Straiker
Editorial guidance for defenders. These are controls an organization can own, not evidence that particular controls were missing or would have prevented this campaign.
Application owners can inventory public services, test access rules and verify that fixes reach every deployed copy.
A patch does not address an already stolen login or every flaw in application logic.
Identity owners can narrow permissions, separate sensitive systems and rehearse revoking credentials and sessions.
Revocation must cover copies and dependent services; it does not undo information already removed.
Responders can investigate suspicious logins, commands and exports together, then preserve evidence and contain affected systems.
An alert requires investigation. Unusual behavior alone does not establish compromise.
Editorial connections to relevant controls, not evidence that a particular technology would have prevented this case. Each guide links to the security guidance behind its recommendations.
Explore the full AI security frameworkExplore more original accounts in the source report library ↗.
These are source-reported findings. An independent assessment, when available, is labeled explicitly.
01Read this case’s update history →
Reviewed Oct 23, 2026 · Editorial methodology · Structured data