AI is making cybercrime cheaper. Hospitals and small banks are the easiest targets
Attacks that once needed a team of skilled hackers can now be run by one person with an AI model. Here is what the numbers show, why local hospitals and community banks are exposed, and the simple defences that still work.

Key points
- Americans reported $20.9 billion in cybercrime losses to the FBI in 2025, five times the 2020 figure. For the first time, the FBI counted AI-linked complaints: 22,364 of them, with $893 million in losses.
- AI helps criminals most with the human part of crime: Microsoft found AI-written phishing emails were clicked 54% of the time, against 12% for traditional ones.
- Anthropic reported in 2025 that one criminal used its Claude Code tool to steal data from and extort at least 17 organisations, including healthcare providers.
- Hospitals are exposed because they can't stop working, hold valuable data and often have tiny security teams. The FBI lists healthcare among the sectors hit hardest by ransomware.
- Deepfakes are now a banking problem: an Arup employee sent $25 million after a video call with a fake CFO, and US regulators have told banks to flag deepfake fraud.
- The defences are unglamorous but effective: phishing-resistant sign-in, offline backups, fast patching and a rule that no money moves on the strength of a call or email alone.
For years, one thing limited cybercrime: there were only so many skilled hackers. That limit is disappearing. “The limit used to be the finite number of malicious hackers,” Michael Kleinman of the Future of Life Institute told The Verge this week. AI tools now let one person do work that, in the words of Anthropic’s head of threat intelligence, used to need “a team of sophisticated actors”.
The organisations least able to cope are the ones people rely on most: local hospitals, community banks and small charities.
The losses keep climbing
The FBI’s Internet Crime Complaint Center publishes the best public count of cybercrime in the US. In 2025 it received more than one million complaints for the first time, and reported losses reached $20.9 billion, up 26% in a year and five times the 2020 figure. The real total is higher, because many victims never report.
Reported cybercrime losses have grown fivefold in five years
Losses reported to the FBI's Internet Crime Complaint Center, billions of US dollars
Source: FBI IC3 Internet Crime Report 2025
The 2025 report is also the first with a section on AI. The FBI counted 22,364 complaints that mentioned AI, with $893 million in losses. Most of that money was lost to investment scams, where AI generates convincing fake advisers, websites and conversations at scale.
Where AI-linked scams cost the most
Losses in complaints that referred to AI, by type of scam, 2025, millions of US dollars. Figures marked 'more than' in the report are shown at their stated minimum.
Source: FBI IC3 Internet Crime Report 2025
What AI actually changes
AI hasn’t invented new crimes. It makes the old ones cheaper, faster and more convincing.
- Better bait. Phishing emails used to give themselves away with clumsy wording. In Microsoft’s 2025 analysis, AI-written phishing emails were clicked 54% of the time, against 12% for traditional ones, four and a half times as effective.
- Less skill needed. In August 2025 Anthropic reported that a single criminal had used its Claude Code tool to break into and extort at least 17 organisations, including healthcare providers and emergency services, with some ransom demands above $500,000. In November it described a likely Chinese state group that used Claude to carry out 80 to 90% of an espionage campaign against about 30 organisations.
- Malware that adapts. Google has found malware that asks an AI model to rewrite its own code every hour so security tools don’t recognise it.
- Fake people. In 2024 an employee of the engineering firm Arup in Hong Kong joined a video call with what looked like the company’s CFO and colleagues. All of them were deepfakes. He made 15 transfers worth about $25 million. US banking regulators have since warned banks about deepfakes used to open accounts and push through payments.
The UK’s National Cyber Security Centre expects AI to make attacks more frequent and to split organisations into two groups: those that keep up, and a larger, more vulnerable group that can’t.
Why hospitals and small banks are exposed
Hospitals are close to an ideal target. They can’t stop working, so they are under pressure to pay quickly. Their records sell well. And many have tiny security teams: a medical centre in Ohio told The Verge it has a single cybersecurity analyst. The FBI lists healthcare among the sectors hit hardest by ransomware; by HIPAA Journal’s reading of its 2025 figures, health organisations filed more ransomware complaints than any other critical sector. In 2024 the ransomware attack on Change Healthcare exposed data on 192.7 million people, the largest health data breach on record in the US.
Community banks and credit unions face a similar gap. Criminals target them with the same tools used against global banks, but they have a fraction of the budget. Their exposure is mainly fraud: fake invoices, cloned voices of executives and deepfaked customers.
There is some help coming. OpenAI has offered $1 billion of subsidised access to its security tools for organisations such as regional banks and public services, and several AI companies now share information on attackers they detect. But the most capable defensive AI tools remain limited to large organisations.
Defences that still work
Almost every attack still depends on a person clicking, a password being stolen, a missing update or a payment made in a hurry. The basics block most of them:
- Phishing-resistant sign-in. Use passkeys or security keys for email and remote access. Codes sent by text can be phished; passkeys can’t.
- Offline backups you have tested. Ransomware is far less damaging if you can restore systems without paying. Keep at least one copy that can’t be changed or deleted from the network.
- Patch fast. AI is shortening the time between a flaw being published and criminals exploiting it. Prioritise anything that faces the internet.
- No money moves on a call or email alone. Confirm payment changes and urgent transfers by calling back on a number you already have, whoever appears to be asking, and however real they sound or look.
- Know who to call. In the US, report fraud to your bank at once and at ic3.gov: the FBI froze $679 million for victims in 2025, and speed makes the difference.
For hospitals, the US health department’s 405(d) programme publishes free, practical security guidance written for small health organisations. It is a good place to start.
Sources
- The Verge: AI is supercharging hacking, and your local hospitals and banks aren't ready (28 September 2026)
- FBI Internet Crime Complaint Center: 2025 Internet Crime Report (April 2026)
- HIPAA Journal: FBI Internet Crime Report 2025, healthcare figures
- Microsoft Digital Defense Report 2025
- Anthropic: Detecting and countering misuse of AI (August 2025)
- Anthropic: Disrupting the first reported AI-orchestrated cyber espionage campaign (November 2025)
- Google Threat Intelligence Group: Threat actor usage of AI tools (November 2025)
- CNN: Arup lost $25 million in Hong Kong deepfake video conference scam (May 2024)
- FinCEN: Alert on fraud schemes involving deepfake media targeting financial institutions (November 2024)
- HIPAA Journal: 2025 healthcare data breach report (February 2026)
- UK NCSC: Impact of AI on cyber threat from now to 2027
- OpenAI: Daybreak for frontline defenders (September 2026)
- US HHS 405(d): Health Industry Cybersecurity Practices
Comments
What do you think? No account needed. Please be kind; comments that attack people are removed.
No comments yet. Be the first.