
What happened
NorForge's Snorre Fagerland scanned exposed Cairn instances and found AIA0001, a Chinese-language operator with 14 panels and over 1,100 attack projects. About 8.4%, or 96 projects, succeeded, targeting e-commerce and payments in the US, Japan and Australia.
Why it matters
Fagerland's finding suggests attackers can direct an AI agent in plain language to hunt weaknesses at scale, and about 8.4% of AIA0001's 96 successful projects hit payment-related targets — a level of automation that may put smaller security teams at a disadvantage.
WHO IT HITSSecurity teams at e-commerce and payment-infrastructure operators in the US, Japan and Australia face automated vulnerability hunting that can succeed in about 23 hours at roughly $50 per successful hack, according to Fagerland's findings.
Summaries like this, in your inbox every morning.
Fagerland's discovery began with a scan for Cairn instances that users had unintentionally left exposed online. From early October 2026 he identified more than 20 malicious panels and watched how the hackers operated, which showed that AIA0001 was running 14 panels and more than 1,100 attack projects in Chinese. His review suggests these were financially motivated actors rather than state-backed intrusion teams.
The scale stands out against the cost. Fagerland says a Cairn-assisted hack succeeded in about 23 hours, and the tokens behind the campaigns cost only around $5,000 in total — roughly $50 per successful breach. Reuters' independent investigation then tied the activity to named manufacturers, and the companies' responses ranged from Deere's remediation to Cummins still investigating a possible third-party website compromise. Cairn's author, who uses the handle le1xia0, did not answer Reuters' questions.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
DeepSeek V4.1 Flash 1017, a 552B-total / 16B-active MoE model with a 1M-token context window and MIT license…

OpenAI said its latest model, GPT-6 Luna, outperformed GPT-5.6 Sol on SQLite-compatible database engine tests…

wevnal released BOTCHAN AICALL on 10月20日

Ricoh built a new Japanese safeguard model (an LLM that detects harmful input and output) on gemma-4-E4B-it, k…

Speak began offering Phone Speak, a phone-style English conversation feature for Premium Plus subscribers (¥29…

Cloudflare announced Clairic, a System One-class model that extracts values and dates from text and images in…
