AI SAFETY
The Rise of AI Hackers: What You Need to Know
Jade Artry
Managing Editor · 26 October 2026
A hacker reportedly used AI agents to break into websites belonging to more than 100 companies, while the automated tools did almost all the work. The operation targeted more than 1,200 sites on an estimated $5,000 AI budget.
It shows how quickly one person can now investigate weaknesses across businesses people shop with and trust. We can’t always see those weaknesses, but the consequences can reach anyone whose information those websites hold. Here’s what you need to know.
Summary
- AI agents targeted more than 1,200 websites and reportedly compromised sites belonging to over 100 companies.
- The extent of customer data theft in this campaign remains unclear.
- Unique passwords, MFA and account alerts help limit the harm from a breach.
How one hacker used AI agents to attack websites
On 19 October, Reuters reported on an investigation into an operation that used Cairn, an open-source system for coordinating AI agents, to probe websites belonging to businesses across the US, Canada, Australia and Japan. Reuters independently examined the activity and identified sites linked to companies including Deere, SharkNinja and Cummins.
Rather than inspecting every website manually, the attacker gave the agents short instructions. They then searched for weaknesses, tested routes into systems and changed approach when an attempt failed. In one case, agents moved beyond a company’s main website to investigate connected infrastructure and eventually found a way to run commands on a server.
The successful intrusions observed in the most active campaign took roughly 23 hours on average. The researcher estimated the AI usage cost at around $50 per successful compromise, although these are estimates rather than independently audited totals.
This is what makes agent-led attacks different from ordinary automated scans. The tools can assess a failed attempt and pursue another line of investigation over many hours. They still have limitations, but they can give a single attacker persistence and reach that previously demanded much more hands-on work.
The concern extends beyond this one campaign. In an October security advisory, India’s national incident response team, CERT-In, described incidents involving AI agents conducting reconnaissance and intrusions against real systems. The weaknesses being exploited are often familiar. What is changing is the speed and scale at which attackers can look for them.
Why a hacked website can become your problem
A successful intrusion doesn’t automatically mean customer information has been stolen. Reuters said it couldn’t establish the extent of any data theft in this campaign. Deere reported no indication that sensitive information was accessed, while SharkNinja said its incident involved a separate product-support website and had been contained.
That distinction matters. But the absence of confirmed data theft in every case shouldn’t make the underlying risk easy to dismiss. Depending on the access gained, a compromised website could expose contact details, customer accounts, order histories or other information, or provide a route into connected systems.
We’ve written before about what happens when personal details become available to people who shouldn’t have them. An email address or phone number can make a phishing message more convincing, especially when a criminal can refer to a real business you use. Even information that seems ordinary can become useful when combined with other details.
Our guide to identity theft monitoring and recovery services explains what protection can and can’t do after information is exposed. No monitoring service can stop a company from being breached, but timely warnings and recovery support can help people respond to misuse.
Smaller businesses have less room for error
For small businesses, this development is particularly difficult. Attackers can keep probing websites around the clock, while a business owner may be relying on a hosting provider, a handful of plugins and an occasional security review. An old administrative account or outdated component can remain unnoticed for months.
That’s why we’ve consistently encouraged businesses to prioritise the basics that prevent common compromises. Our guide to securing a small business without a dedicated IT team covers practical steps, including keeping systems updated, protecting administrator access and reviewing the services connected to a website.
Even when no customer records are stolen, an intrusion can interrupt sales, require expensive recovery work and damage customer confidence. Businesses that collect personal information are responsible for protecting it, regardless of how sophisticated an attacker’s tools become.
AI agents are also raising questions about bank account access
A separate Reuters report on 22 October said major US banks were blocking unvetted third-party AI agents attempting to operate inside customer accounts. In some cases, customers had shared their own banking credentials with an agent. This was not part of the website-hacking campaign, and the reporting did not establish widespread theft from customer accounts.
It does, however, highlight another emerging question. When an AI tool asks to connect to a sensitive account, what can it do once access is granted, and how easily can that access be removed? For financial accounts, we’d use only clearly authorised integrations and keep login credentials out of third-party AI chats and agents.
What we recommend doing now
There’s no reason to assume every website is compromised, and people can’t personally patch the systems of every company they use. But there are sensible ways to reduce the harm if a service experiences a breach.
- Make account access harder to abuse. Use unique passwords, turn on multi-factor authentication and choose passkeys where they’re supported. A stolen password shouldn’t unlock several accounts.
- Be cautious about connected AI tools. Check what an agent can access before authorising it. Avoid giving passwords or security codes to services that aren’t explicitly supported by the account provider.
- Turn on financial and account alerts. Notifications for unfamiliar sign-ins, changed account details or transactions can help you respond sooner.
- Treat breach notices seriously. Check the company’s official website or app rather than following unexpected links. Secure any affected accounts and watch for convincing follow-up scams.
- Review what information you leave with services. Close unused accounts where practical and avoid supplying optional personal details a business doesn’t need.
For business owners, our advice is to start with internet-facing systems and the accounts that control them. Remove unused software and accounts, apply security updates, require multi-factor authentication for administrators, maintain reliable backups and check who can access customer data.
Credentials deserve particular attention when employees, contractors and external providers share access. We’ve covered secure password storage for businesses because an exposed password or API key can give an attacker a shortcut past other defences.
As AI grows more capable, we are likely to see more of this
The UK’s National Cyber Security Centre has assessed the threat through 2027 and expects AI to make parts of cyber intrusions faster and more effective. It also cautions that fully automated, advanced attacks are unlikely to become the norm immediately. Human attackers remain part of the picture.
That’s the balance worth keeping in mind. AI can help defenders find weaknesses too, and not every attempted intrusion succeeds. But the cost of testing thousands of potential weaknesses is falling, while many organisations are still struggling with the same preventable security gaps.
We expect more incidents in which AI agents do much of the work. The responsibility for securing customer information remains with the organisations collecting it. For the rest of us, staying informed, protecting accounts and knowing how to respond when something goes wrong are becoming increasingly important.
Sources and further reading
- Reuters investigation into AI-agent website intrusions, 19 October 2026
- CERT-In advisory on AI agents as attack vectors, 15 October 2026
- Reuters reporting on third-party AI agents and bank accounts, 22 October 2026
- NCSC assessment of AI and cyber threats through 2027
