Closing Time: One hacker, an army of AI agents

A single hacker no longer needs a crew. A Chinese-speaking cybercriminal broke into websites belonging to more than 100 companies, including farm equipment maker Deere, appliance company SharkNinja and engine manufacturer Cummins, while doing almost none of the work, Reuters reports in an exclusive. AI agents spent hours autonomously probing the targets and breaking in, according to Snorre Fagerland, principal threat researcher at the cybersecurity company NorForge.

Fagerland was able to watch it happen. While scanning the web for exposed instances of Cairn, an open-source AI problem-solving platform popular with hackers, he found about two dozen being used for malicious activity beginning around October 2, which let him quietly look over the criminals’ shoulders. One of them stole a trove of credit card data, and another tried to get AI agents to rob a cryptocurrency exchange.

The most prolific hacker barely had to give directions. Their instructions sketched out preferred intrusion tactics in fewer than 100 Chinese characters, and starting on September 25 their agents tried to break into more than 1,200 e-commerce sites in the US, Canada, Australia and Japan. Successful intrusions typically took around 23 hours, and Fagerland estimated the campaign, which appeared to use a mix of DeepSeek and Anthropic’s Claude Code, used roughly $5,000 worth of tokens, or about $50 per successful hack.

The agents were not just running simple scripts. In one case, after failing to find any weakness in a large company’s public-facing application, they searched digital records for related infrastructure, found a public-facing control panel and, after hours of trial and error, reported tricking the server into executing malicious commands. “It’s not script kiddie stuff,” Fagerland said.

The companies say the damage was limited. Deere said it had taken “necessary remediation actions” and had “no indication that sensitive information was accessed,” SharkNinja said the attack hit a standalone product-support site and was contained, and Cummins said it was investigating a potential attack on a site run by a third-party vendor. Reuters could not determine how much data, if any, was taken.

What changed is how little supervision the work now requires. Fagerland said he had assumed hacker agents would still need some human guidance, and that it is now obvious the technology has moved past that. He called the find real-life evidence that commercially available AI agents can carry out complex intrusions from start to finish without human intervention, a dramatic boost to the power of a single attacker.

Elsewhere in AI today:

Reuters reports Amazon is introducing TorchNeuron, a free, open-source tool that lets its Trainium AI chips run code written in PyTorch with a single configuration change and no rewriting. The move takes aim at the software advantage behind Nvidia’s dominance, and Amazon’s Annapurna Labs says more than 30 early customers in robotics, world models and voice AI are already using it.

The Verge reports the Department of Homeland Security ordered US civilian agencies to remove any technology made by DeepSeek over concerns its models may transfer “sensitive data to the PRC.” The ban also bars agencies from using third-party services that build in DeepSeek’s models.

TechCrunch reports attackers broke into an older system at OpenCode, the open-source coding agent, affecting 1.4% of users of its OpenCode Console, where some developers store their AI API keys. Users reported unauthorized charges in the hundreds of dollars, with a few above $1,000, and the team says AI models it used for analysis started refusing to help once the malicious code tripped their safety filters.

TechCrunch reports inference chip startup Tenstorrent raised $1.5 billion at a $6.2 billion valuation, more than doubling its value from late 2024. The round comes five months after Intel and Qualcomm reportedly showed acquisition interest, and the money will go toward chips built for serving open-weight models and video generation.

CNBC reports traders on Kalshi now see Anthropic as the only major company likely to officially announce an IPO this year, with 75% odds as of Monday morning. OpenAI sits at 3% after Sam Altman ruled out a 2026 listing, and the number of US IPOs priced this year is down 33% from 2025, according to Renaissance Capital.

---

Markets

Chipmakers led a rebound on Monday as dip buyers bet the AI investment boom will keep supporting solid earnings, Bloomberg reports, with Meta’s results this week the next test for the trade. The Nasdaq rose about 1.1% and the S&P 500 gained about 0.4%, while the Dow finished roughly flat. 62 of the 86 AI and tech stocks we track closed higher.

Storage names did the heavy lifting. Western Digital climbed 6.1% and Sandisk 5.5%, and Amazon rose 5.0% on the day it took aim at Nvidia’s software moat. Micron added 3.0%, Intel 2.4%, AMD 1.8% and Nvidia 1.4%. CoreWeave was the exception: despite a Goldman Sachs upgrade to buy, noted by Yahoo Finance, it gave back its early gains and closed down 0.4%.

Biggest winners

Biggest losers

Prices are Monday’s U.S. cash close, October 19, 2026 (vs. Friday’s close). Market data is a snapshot, not live.

---

Deals Desk

Two compute deals worth knowing.

Anthropic × Nscale — Oct 19, 2026

Computing power for a second facility at Nscale’s Monarch data center campus · West Virginia, US

Potential deal worth about $45 billion; Anthropic already has Nscale agreements worth as much as about $44.6 billion, according to Nscale’s IPO filing

Status: In talks (reported) · Source: Bloomberg

Aolani × Malaysian data center sites — Oct 19, 2026

Letters of intent to take up AI data center capacity across multiple Tier III sites · Malaysia

500 MW, with operations targeted for the second half of 2027; brings Aolani’s planned Malaysia capacity to as much as 580 MW by 2027

Status: Announced · Source: Xinhua

Full tracker → https://www.theaigentic.com/deals