A year of attacks on hospitals, water plants and websites
If it feels like the United States has been under attack all year, the numbers agree. Americans filed more than one million cybercrime complaints with the FBI in 2025, the first time the count has crossed that mark, and reported losses of almost $21 billion. For the first time in 25 years the FBI's annual report had a section on artificial intelligence, linking more than 22,000 complaints and nearly $893 million in losses to AI-enabled fraud.
2026 has kept the pace. What makes this year different is not one giant hack but the spread of targets: hospital clinics, municipal water systems, industrial freezers, and the trusted scripts that run inside some of the world's biggest websites. Here is what actually happened, in order.
Ransomware hits the University of Mississippi Medical Center and takes down its electronic medical records. Roughly three dozen clinics across Mississippi close and elective procedures, including some chemotherapy appointments, are cancelled. Emergency care keeps running while staff document by hand, and the clinics reopen after more than a week.
After the war with Iran begins on 28 February, US agencies warn in April that Iran-affiliated hackers are disrupting programmable logic controllers, the small computers that run pumps and valves, at water, energy and local government sites. A July update adds Schneider Electric and Siemens equipment to the list, days before several Minnesota water utilities report suspicious activity. CISA then warns of a significant increase in attacks on the water sector.
Cedar County Memorial Hospital in Missouri discloses a cyberattack that disables operations for about two weeks and stops card payments. It later confirms data was stolen. A newer ransomware group claims the attack and lists the hospital on its leak site.
CISA, the FBI and the Departments of Agriculture, Health and Human Services, and Transportation issue a joint warning about attacks on internet-connected refrigeration systems. Supermarket freezers abroad were warmed until food spoiled, the Department of Defense is investigating freezers at US military bases, and a blood bank in France caught an attack in time. The government has not named who is behind it.
Attackers exploit a flaw in Optimizely, a popular website testing tool, and plant malicious JavaScript in customer projects. For roughly 17 hours visitors to 178 organisations' sites, including Zoom, the BBC, CNN, HSBC and IKEA, are served code that skims card details, swaps cryptocurrency wallet addresses and targets unpatched iPhones. None of those companies had their own servers broken into.
Four patterns behind the headlines
Strip away the company names and the same four weaknesses show up again and again.
Ransomware on services people need
Hospitals, libraries and city systems are targeted because downtime hurts immediately, which raises the pressure to pay. Many groups now steal data first and threaten to publish it, so a good backup alone no longer ends the problem.
SEEN IN: UMMC, CEDAR COUNTY MEMORIALMachines left open to the internet
Controllers and freezers were reachable from the internet, sometimes with weak passwords or on equipment the vendor no longer patches. The government's advice was basic: take them offline, change passwords, patch, and retire unsupported kit.
SEEN IN: WATER UTILITIES, COLD CHAINTrusted vendors as the way in
Why break into 178 companies when one supplier's script runs on all of them? Every marketing tag, analytics snippet and software package a company loads is a door that someone else holds the key to.
SEEN IN: OPTIMIZELY, 16 NOVEMBERAI-written scams at scale
Criminals use AI to write convincing phishing in any language, clone voices and fake documents. IBM found one in four malicious breaches in its 2026 study were AI-enabled, and the FBI now tracks AI fraud as its own category.
SEEN IN: FRAUD, PHISHING, DEEPFAKESWhy human teams alone cannot keep up
Defenders are not losing because they are careless. They are losing on volume and speed. Close to 3,000 cybercrime complaints reach the FBI every day. More than 100,000 new software vulnerabilities were published between January and 18 November 2026, about 2.4 times the number in the same period of 2025, according to one count of the US National Vulnerability Database. And IBM's 2026 study found that the average breach still took 247 days to identify and contain.
The twist is that only a tiny fraction of those vulnerabilities are ever used by attackers: Google observed exploitation of just 0.23 percent of those disclosed in 2026 through August. The hard part is not patching everything. It is knowing, today, which few matter on your network. That is a sorting problem, and sorting at machine speed is exactly what AI is good at.
How AI actually stops attacks
"AI security" is used loosely, so here is what it concretely does inside a security operations centre (SOC), and how each one maps to the attacks above.
Spots behaviour that does not fit
Models learn what normal looks like for each user, server and machine, then flag what does not: a contractor account logging in from a new country at 2 AM, thousands of files renamed in minutes, a controller receiving commands from an unknown address. This is how ransomware gets caught before encryption spreads.
STOPS: RANSOMWARE, ACCOUNT TAKEOVERTriages alerts in seconds
A SOC can receive thousands of alerts a day, most of them harmless. AI agents gather the evidence for each one, close the obvious false alarms and hand analysts a short list with the context already attached. In a Cloud Security Alliance study, analysts working with an AI agent finished investigations 45 to 61 percent faster, and were more accurate in both test scenarios.
STOPS: ALERT FATIGUE, MISSED SIGNALSRanks which holes to patch first
Instead of a list of 100,000 vulnerabilities, AI combines exploit chatter, what is actually exposed to the internet and how critical each machine is, and produces a short patch queue. The water and freezer attacks were the textbook case: exposed, unpatched, and findable.
STOPS: EXPOSED DEVICES, OLD FLAWSWatches what third-party code does
Browser-side monitoring tools learn what each vendor script normally touches. When a trusted testing script suddenly starts reading card fields or rewriting wallet addresses, as on 16 November, that change in behaviour is the alarm, even though the file still comes from the vendor's own servers.
STOPS: SUPPLY-CHAIN ATTACKS, SKIMMERSReads phishing the way an attacker writes it
Language models score emails and messages on intent, not just on known bad links, so a perfectly written fake invoice or an urgent request from a cloned CEO voice still gets flagged. Banks run similar models on payments to stop fraud before money leaves.
STOPS: PHISHING, DEEPFAKE FRAUDContains damage while people wake up
With approved playbooks, AI can isolate an infected laptop, disable a compromised account or block an address in seconds, then open the incident ticket. IBM found companies using AI and automation extensively in security saved almost $2 million per breach on average.
STOPS: SPREAD, DOWNTIME, COSTA ransomware alarm at 2 AM, with AI in the loop
To make it concrete, here is how a modern hospital SOC would handle the start of an attack like the one in Mississippi.
An IT contractor's account signs in through the remote access gateway from a country the contractor has never logged in from. On its own, a human might not look at this until morning.
The detection model links that sign-in to three file servers where thousands of files are being renamed with a new extension. It scores the case as likely ransomware and pulls the evidence into one view.
Following a playbook the hospital approved in advance, the system disables the account, cuts the three servers off the network and raises a priority one incident in the ticketing tool.
The on-call analyst confirms it is real. The major incident manager opens a bridge call with infrastructure, the security vendor and the hospital's IT head, sets the update rhythm and starts the communication to leadership.
Clean servers are restored from backup, the root cause (a stolen contractor password with no second factor) goes to problem management, and patients never notice. Without the AI step, the first sign might have been nurses unable to open records at shift change.
What this means for careers in India
Every attack in this article ended with people on a call: analysts confirming the alert, engineers isolating systems, an incident manager running the bridge, and a problem manager making sure it cannot happen the same way twice. AI is changing how much of the legwork they do, not whether they are needed. Much of that round-the-clock monitoring and incident work for global companies is done from India, by IT services firms, managed security providers and global capability centres.
Demand reflects it. TeamLease Digital estimates India has around four to five lakh cybersecurity professionals while employers are trying to fill nearly 40,000 open positions, and that demand for specialists rose 15 to 20 percent between 2025 and 2026. In ISC2's September 2026 survey of Indian practitioners, AI was the most in-demand skill, named by 52 percent, ahead of cloud security.
Incident and major incident management
The people who take over at 02:25: they run the response, coordinate teams, keep leadership informed and track it to closure within the SLA. No programming needed, strong process and communication skills are.
TRAINED IN: OUR STOM PROGRAMBuilding the AI that detects
Data scientists and AI engineers build the anomaly models, alert-scoring systems and agents described above. Python, machine learning and the habit of testing models against real data are the core skills.
TRAINED IN: DATA SCIENCE AND AILearn to run the response when it matters
Our STOM program covers ITIL 4, incident and major incident management, problem management and root cause analysis, change and release, and hands-on practice in ServiceNow, Jira Service Management and BMC Remedy. Open to any graduate, no programming background needed, with placement support until you are hired.
See the STOM programQuestions people ask
Were the big US outages in November 2026 cyberattacks?
As of 19 November 2026, none of the six major outages between 12 and 17 November (Microsoft 365, the FAA, JPMorgan Chase, United Airlines, the NSA and Apple) had been attributed to a cyberattack by the organisation affected, and US officials reported no sign of one. Attackers did run phishing campaigns using the outages as bait.
Can AI stop cyber attacks on its own?
No. AI is very good at spotting unusual behaviour, sorting alerts and taking fast first actions like isolating a machine. People still confirm what is real, decide on bigger actions, coordinate the response and fix the root cause. Attackers also use AI, so the advantage goes to teams that combine both.
Do I need to code to work in cyber incident response?
Not for incident management, major incident management or problem management roles, which are about running the response process, coordinating teams and communicating clearly. Coding matters more for security engineering and for building AI detection models.
Which skills are employers asking for in 2026?
In ISC2's September 2026 India findings, AI skills were the most in demand at 52 percent, followed by cloud security, application security, security analysis and risk management. Tool experience with ITSM platforms such as ServiceNow is also common in incident and operations job descriptions.
Is cybersecurity a good career in India right now?
Demand is growing faster than supply. TeamLease Digital estimates nearly 40,000 open positions and a 15 to 20 percent rise in demand between 2025 and 2026. Entry points include SOC analyst, incident management and security operations roles, with growth into specialist and leadership positions.
Sources
FBI IC3 2025 Internet Crime Report: 1,008,597 complaints, $20.877B in losses, up 26%, first AI section with 22,364 complaints and about $893M in losses (Apr 2026), via helpnetsecurity.com, infosecurity-magazine.com and theblock.co
IBM Cost of a Data Breach 2026 (breaches March 2025 to February 2026): US average $11.5M, global $6M, 247-day lifecycle, almost $2M saved with security AI and automation, one in four malicious breaches AI-enabled, one in five breaches targeting AI models, shadow AI incidents 43%: csoonline.com
University of Mississippi Medical Center ransomware (Feb 2026): securityweek.com, esecurityplanet.com, techtarget.com
Iran-affiliated PLC attacks on US critical infrastructure, April advisory and July update: techcrunch.com, securityaffairs.com; Minnesota water utilities and CISA water sector alert (Jul 2026): route-fifty.com, eenews.net
Cedar County Memorial Hospital ransomware (Aug to Sep 2026): paubox.com
Joint US government warning on industrial refrigeration hacks (5 Nov 2026): techcrunch.com
Optimizely supply-chain attack, 178 organisations, about 17-hour window (16 Nov 2026): cmswire.com, techaeris.com, decryptiondigest.com
November 2026 outages and their stated causes, NVD vulnerability counts and Google exploitation share: stingrai.io
Cloud Security Alliance, Beyond the Hype: A Benchmark Study of AI Agents in the SOC (Oct 2025), 148 participants: cloudsecurityalliance.org, virtualizationreview.com
India demand: TeamLease Digital estimates, via sightsinplus.com; ISC2 September 2026 India findings, via masaischool.com
The 2 AM hospital scenario above is an illustrative composite of a common ransomware pattern, not a specific incident.