Windows malware sends stolen data to Claude to rank victim assets then self-deletes
Tom's Hardware portrays new Windows tooling that exfiltrates victim data to Claude for asset ranking before self-deletion, with fallbacks that swap models or omit AI when blocked. The Decoder reported a tampered Claude Code extension on Cursor's marketplace for roughly eleven hours, affecting up to five thousand users until Cursor shut the store down, while TechCrunch said GMO Internet is verifying alleged customer data after suspected Onamae cloud-desktop access where twenty-seven thousand servers may be involved amid unauthorized trading reports. Anthropic added Claude for Legal templates with more than sixty integrations. The convergence stresses API misuse on endpoints, poisoned extensions, and trading-cloud exposure for defenders and providers. Tom's Hardware excerpts do not establish malware prevalence, the precise Claude variant, or independent validation of the implant.
Windows malware sends stolen data to Claude to rank victim assets then self-deletes
Tom's Hardware reports new AI-driven Windows malware that feeds stolen data to Claude to rank victims' assets by value, then deletes itself. The payload builds custom target assessment and can switch models if one refuses or skip the AI step entirely.
Key takeaway
Commercial LLMs are being wired into malware workflows for automated victim triage, with fallbacks that weaken refusal-only defenses.
What happened
Tom's Hardware reports new AI-driven Windows malware that feeds stolen data to Claude to rank victims' assets by value, then deletes itself after completing that assessment.
The same reporting says the payload builds a custom target assessment, can switch models if one refuses, or skip the AI step entirely; bundled items also describe a malicious Claude Code extension on Cursor's marketplace for about eleven hours affecting up to five thousand users and GMO Internet's investigation of a suspected Onamae cloud desktop breach.
Evidence
Windows malware reportedly sends stolen data to Claude to rank victim assets by value, then self-deletes.
Tom's Hardware AI · attributed
Tom's Hardware reports new AI-driven Windows malware that feeds stolen data to Claude to rank victims' assets by value, then deletes itself.
The malware payload can switch models if one refuses or skip the AI step entirely.
Tom's Hardware AI · attributed
The payload builds custom target assessment and can switch models if one refuses or skip the AI step entirely.
A tampered Claude Code extension was live on the Cursor marketplace for about eleven hours and affected up to five thousand users.
The Decoder · attributed
A tampered Claude Code extension was live on the Cursor marketplace for about 11 hours, affecting up to 5,000 users.
GMO Internet is investigating suspected unauthorized access to Onamae cloud desktop service with up to twenty-seven thousand servers potentially affected.
TechCrunch AI · attributed
GMO Internet is investigating a data leak after suspected unauthorized access to its Onamae cloud desktop service, which provides virtualized desktops to financial traders.
Anthropic is adding pre-built templates to Claude for Legal with more than sixty integrations.
The Decoder · attributed
Anthropic is adding pre-built templates to Claude for Legal. These interactive overviews and tables simplify setup for legal teams and speed up routine work like contract reviews, while more than 60 integrations let firms connect their own databases directly.
Why it matters
Security teams and model providers face overlapping abuse paths: API-driven triage in endpoint malware, compromised IDE extensions, and high-value cloud desktop breaches in the same news cycle.
Limits and uncertainties
The Tom's Hardware coverage is excerpt-only and does not establish deployment scale, the specific Claude model, or independent verification of the Windows malware.
GMO has not publicly linked the Onamae breach to the Artex penetration-testing tool mentioned in related reporting, and the full extent of that incident remains unclear.
Practical implications
Endpoint and SOC teams should treat unexpected outbound LLM API traffic from Windows hosts as a high-priority hunting signal alongside traditional exfiltration patterns.
IDE marketplace operators should tighten extension review and incident rollback playbooks given the reported eleven-hour window affecting thousands of users.
What to watch
Follow-on technical write-ups naming samples, C2 endpoints, or confirmed Claude API usage for the reported Windows malware.
GMO customer advisories confirming verified leak contents and scope beyond the twenty-seven thousand server exposure estimate.
Original reporting: New AI-driven Windows malware feeds stolen data to Claude to rank victims’ assets by value, then deletes itself — payload builds custom target assessment, can switch models if one refuses, or skip the AI step entirely - Tom's Hardware
