Hacker used AI agents to breach John Deere, SharkNinja, Cummins and 100+ sites

A Chinese-speaking hacker used largely autonomous AI agents to break into websites of more than 100 companies, including Deere, SharkNinja and Cummins, Reuters reported on 19 October 2026 [2]. Security firm NorForge spotted the campaign through unsecured control panels of Cairn, an open-source agent platform [1][2].

How did NorForge find the hacker?

NorForge found more than 20 exposed Cairn panels that it says were likely used for malicious hacking. Principal threat researcher Snorre Fagerland began finding them around 2 October by scanning the web, according to Reuters [1][2].

Cairn describes itself as a general-purpose problem-solving engine that searches for a path from an origin to a goal, with penetration testing as its first validated domain [3]. Agents post facts and intents to a shared board, and a human operator can add hints [1][3].

The open panels exposed targets, findings and credentials, NorForge says. Its Trufflehog scan found hundreds of secrets, 11% of which were verified as valid [1].

What did the AI agents do?

NorForge calls the most active actor cluster AIA0001, with over 1100 attack projects across 14 dashboards, mostly against e-commerce sites in the US, Japan and Australia [1]. Reuters reports the campaign began on 25 September and had tried to break into more than 1,200 sites by late the previous week [2].

Successful intrusions took a little under 23 hours on average, according to NorForge [1]. The tasks it saw were run by a worker named claudecode_deepseek-v4.1-flash, which NorForge says could mean a Claude Code harness running DeepSeek's 4.1 Flash model [1].

Snorre Fagerland estimated the campaign used about $5,000 of AI tokens, or about $50 per successful hack. NorForge's blog puts the average cost per task at around $5 and warns its numbers could be off by a clear margin [1][2].

In one case, the agents found no weakness in a large company's public application, then located a related control panel and, after hours of trial and error, reported tricking the server into running commands, according to Reuters [2].

What did Deere, SharkNinja and Cummins say?

Deere said it had no indication that sensitive information was accessed after the incident. The company told Reuters it had taken necessary remediation actions [2].

SharkNinja said the hack hit a standalone product-support site and that the threat was contained [2]. Cummins said it was investigating a report of a potential attack on a website run by a third-party vendor [2][4].

NorForge's report does not name the victims; Reuters said it identified them by independently reviewing the hacker's Cairn activity [2]. Reuters could not establish how far the breaches led to data being taken [2].

Cairn's author, who posts as le1xia0 on X, wrote on 8 October that the platform was not built for attacks [5].

The AI Pin has covered rogue agents released by AI labs, such as OpenAI's disclosure of four more incidents (theaipin.com/drama/openai-discloses-four-more-rogue-agent-incidents). This case is different, because a human criminal aimed the agents on purpose.

Our read

Opinion: The striking number is not 100 breaches but $50 each. Defenders should assume that patient, cheap, automated probing of every public endpoint is now normal, and that attackers' own sloppy tooling can leak victims' secrets even further.

What we could not verify

• The victims' names come from Reuters' own review of the Cairn data; NorForge does not name them, and we could not see the panels.
• The $5,000 and $50 figures are Fagerland's estimates; NorForge says it has no actual token usage numbers.
• Whether data was taken; Reuters could not ascertain how far the breaches led to data extraction.
• Which models were used; the worker name suggests Claude Code with DeepSeek 4.1 Flash, but that is NorForge's inference.
• Reuters blocked our direct fetch; we checked its text in a syndicated copy on CNBC-TV18.
• The Deere and SharkNinja statements reached us only via Reuters; neither had answered The Verge.