Case sysdig-llmjacking-deepseekLLMjackingWorldwide
LLMjacking crews load stolen DeepSeek keys into reverse proxies days after model launch
Sysdig found over 20 reverse proxies running on stolen credentials; one held 55 DeepSeek keys, another burned 2.2 billion tokens worth tens of thousands of dollars in 4.5 days and resold access for 30 USD a month.
HighAttack on AIStatus: OngoingVerified case file
Open in the interactive mapCite this case
- Reported
- Feb 7, 2025
- Location
- Worldwide
- Target
- Cloud and AI API account holders (OpenAI, AWS, Azure)
- Severity
- High
- Triggered by
- Unknown
- Actor type
- Criminal
- Reported loss
- None reported
- CVE
- None
What happened
Sysdig found over 20 reverse proxies running on stolen credentials; one held 55 DeepSeek keys, another burned 2.2 billion tokens worth tens of thousands of dollars in 4.5 days and resold access for 30 USD a month.
Impact: One sampled proxy showed 132,651 prompts and 2.213 billion tokens, estimated at roughly 49,596 USD of victim spend in 4.5 days; Claude 3 Opus alone accounted for an estimated 38,951 USD.
Who triggered it
Actor: Unknown
Attribution: Sysdig TRT observed operators sharing access on Rentry and 4chan and hiding behind TryCloudflare tunnels; individuals not identified.
Motive: Avoiding LLM costs and reselling model access
How it happened
Attack vector: Credentials stolen from vulnerable services such as Laravel or scraped from public repositories, verified by scripts and loaded into ORP proxies
Root cause: Leaked long-lived API keys with no usage caps; proxies made abuse look like legitimate traffic
Timeline
- DeepSeek-V3 released; support appears in an ORP within days
- DeepSeek-R1 added to proxies the day after its release
- Sysdig publishes findings on DeepSeek-focused LLMjacking
Response
Sysdig published indicators and urged key rotation, usage limits and secret scanning; affected providers were notified.
Current status: Ongoing
Sources (open in a new tab)
- Sysdig sysdig.com
- SecurityBrief securitybrief.co.uk
Cite this case
LLMjacking crews load stolen DeepSeek keys into reverse proxies days after model launch. AI Threat Watch, case sysdig-llmjacking-deepseek, Feb 7, 2025. https://aithreat.io/cases/sysdig-llmjacking-deepseek/
Permanent link: https://aithreat.io/cases/sysdig-llmjacking-deepseek/