The World Under Cyber Attack: Major Hacks, AI-Powered Threats and How to Protect Yourself

Every few years a report lands that changes how security officials talk. This is one of those years. In late June, the heads of the Five Eyes cyber agencies, including Canada's Cyber Centre, wrote that frontier AI is reshaping cyber risk and that "the timeline is not years, it is months".
Since then, the headlines have only gotten louder: the largest crypto theft on record, supermarket freezers sabotaged in France, water systems tampered with in the United States, a Canadian charity drained in half an hour, and the first reports of AI systems running parts of an attack on their own.
This is our country-by-country guide to the major attacks of 2025–2026, how AI is changing them, what you can do about it, and what could happen if things get out of hand. It is meant to make you careful, not afraid. Almost everything below can be made much harder for attackers with habits that cost little or nothing.
The short version
Attacks are getting faster. Microsoft measured under three hours from first targeting to privileged cloud access in some cases involving an AI framework, according to Canada's Cyber Centre.
AI is real, but it is mostly an accelerator. Europe's cyber agency says attackers mostly use consumer-grade AI to augment existing skills. Phishing, stolen passwords, exposed systems and software flaws still do most of the damage.
The weak point is usually someone else's system. A backup tool, a support platform, a website-testing script, a contractor. In 2026 the way in was a trusted third party over and over.
Attacks now reach the physical world. Refrigeration controllers, water-system controllers and rail radio signals all show up in this year's incident reports.
The basics still work. Phishing-resistant sign-in, fast updates, tested backups and a habit of verifying unexpected requests blunt most of what you will read below.
How we sourced this. Every item links to its source: a government agency, the affected organization, a court or regulator filing, or established news outlets. Figures are as reported by the named source. Claims made on criminal leak sites are labelled unverified, because criminals lie. This is a fast-moving story and numbers change as investigations continue. Stingrai's running timeline of 72 confirmed 2026 incidents, each linked to a primary source, helped us find several of the items below.
Canada: lead story, closer to home than you think
Canada's Cyber Centre said in its National Cyber Threat Assessment 2025–2026 that state adversaries are becoming more aggressive, that AI is "almost certainly lowering the barriers to entry" for attackers, and that adversaries "very likely consider civilian critical infrastructure to be a legitimate target for cyber sabotage in the event of a military conflict." Its ransomware outlook adds that all Canadian organizations, regardless of size or sector, are at risk.
In late October, the Cyber Centre went further and published a bulletin on agentic AI. It says cyber attacks are not yet fully autonomous, and that humans still choose targets and objectives. But it warns that more of the work between those decisions can now be handed to machines, which "gives defenders less time to detect and stop an attack."
Alberta
Bragg Creek Community Association (confirmed). On Nov. 5, Interac warned the association about suspicious transactions. CBC reports that three accounts were drained of hundreds of thousands of dollars in roughly 30 minutes, at about one transaction a minute, with the money sent to another financial institution. The fraudulent transfers used the email address of a local business the association works with. That business received a notice after each one.
The RCMP is investigating, the bank is conducting its own investigation, and the association's internal IT review found the accounts were attacked from outside. The group is trying to raise $50,000 by Nov. 20 to cover about $30,000 in bills and keep the community centre open. Its president says there is no guarantee the money can be recovered.
Mount Royal University, Calgary (confirmed). The university confirmed in July that a ransomware actor took folders from the shared drive used by students and employees, then deleted the data. It is offering two years of credit monitoring and says it will not comment on whether it paid a ransom.
Leduc County (confirmed). The county said it learned on Dec. 25, 2025 that it had been hit by a ransomware attack that disabled some IT systems. Early forensic work found minimal information affected.
Across Canada
Optimizely supply-chain attack (Nov. 16). Attackers injected malicious code through Optimizely Web, a service sites use to test page variations. Security firm Report URI says it found infected code in 178 of the 1,611 snippets it checked, and that the real number is likely higher. According to The Circuit's report, the script swapped crypto wallet addresses, skimmed card numbers from about 8:40 a.m. ET and, from about 12:15 p.m. ET, loaded iPhone and iPad exploit kits aimed at older iOS versions.
Report URI named Equifax Canada, CTV News, TSN, Crave, BNN Bloomberg, CP24 and Starbucks Canada among affected sites. That list comes from Report URI. None of the companies had confirmed it when The Circuit published. Optimizely says the hole was in its account sign-up flow, that it is fixed, and that affected customers have been notified. It is not yet clear whether any devices were actually compromised.
Law firms and fake IT calls (October). At least four of Canada's largest law firms were breached after staff received calls and Teams messages from people pretending to be internal IT. The Globe and Mail reports that employees were sent to fake sign-in pages that bypassed multi-factor authentication. McCarthy Tétrault and BLG confirmed unauthorized access to files including client information, and Davies reported one compromised account. Bennett Jones said its document-management and accounting systems were unaffected. A source told the Globe that 31,000 files were stolen at McCarthy Tétrault; the firm has not confirmed a number.
The Cyber Centre issued an alert (AL26-028) on the same campaign, which it says has hit legal and financial organizations and uses real-time relay to get around ordinary MFA. Its advice: phishing-resistant MFA such as security keys or passkeys, and verify any unexpected IT call independently, "regardless of caller ID or urgency."
Crime Stoppers tip platform (confirmed). Hacktivists stole data from Navigate360's P3 tip-management platform. CBC analyzed a copy and found more than 1.5 million tips submitted to Canadian Crime Stoppers programs over 25 years, along with tens of thousands of unredacted personal details. Navigate360 acknowledged a breach between November 2025 and March 2026. Privacy law professor Michael Geist called it "potentially one of the most harmful data breaches we've ever seen in Canada." The federal Privacy Commissioner is investigating.
Burnaby School District (confirmed breach, unverified leak). The district says attackers got in on Sept. 21 and copied files containing personal information. CBC found screenshots on the leak site of the Qilin ransomware group, which claims to be offering stolen district data. The district has hired an outside expert to determine whether its data was actually posted. Treat Qilin's claim as unverified until the district confirms.
Loblaw (confirmed). A criminal used compromised credentials for a quality-assurance platform to get into Loblaw systems. According to the Privacy Commissioner's compliance letter, about two million Canadians had details such as names, contact details, birth dates and loyalty numbers exposed, plus partial payment-card digits.
Investment regulator and driver's licences. The Canadian Investment Regulatory Organization said in January that a phishing attack affected about 750,000 investors, including social insurance numbers, per the Stingrai timeline. And an ID-verification company, IDScan, confirmed hackers reached customer data after a service was found selling what it claimed were 150 million-plus US and Canadian driver's licence records.
The money side. The Canadian Anti-Fraud Centre says Canadians reported about $351 million in fraud losses in the first six months of 2026, and that only 5% to 10% of fraud is reported. Spear phishing losses reached $57.7 million in six months, compared with more than $68 million for all of 2025.
United States
The US saw some of the year's most disruptive incidents, and many were aimed at the physical systems that keep daily life running.
Stryker (March). The medical-device maker said it suffered a global disruption to its Microsoft environment, and employees reported devices wiped. A group called Handala claimed responsibility and said it wiped more than 200,000 devices. That figure is the group's claim. Stryker later said the incident had a material impact on operations.
Water and industrial controllers (April–July). Six US agencies said in advisory AA26-097A that Iran-affiliated actors were exploiting internet-exposed industrial controllers across several sectors. In July, Minnesota reported a coordinated attack on more than 30 community water systems; the state said water quality was not affected. Minnesota has not attributed the attack.
Cisco SD-WAN (February). CISA issued Emergency Directive 26-03 after exploitation of Cisco SD-WAN systems posed an imminent threat to federal networks.
Bankruptcy courts (October). A cyber incident at a contractor, BAE Systems, took down the national bankruptcy noticing center. The courts declared a Bankruptcy Rules emergency, and the center was back online by Oct. 27.
Health and education data. DentaQuest reported that 15 million people were affected after a May intrusion. Instructure, which runs the Canvas learning platform, reported data theft affecting more than 30 million students and staff, according to TechCrunch's roundup. Vanderbilt University restricted most systems on Nov. 9 and says full restoration may take weeks.
The ShinyHunters phone-call wave. Jack Henry, a banking-technology provider, said an incident began with a phone call from the group, the same style of fake-IT vishing seen in Canada. It says fewer than 10 client institutions had personal data affected and that it is not paying.
On state activity, CISA's earlier advisory on Volt Typhoon warned that China-linked actors were maintaining persistent access to US critical infrastructure. In March, the US Justice Department said Iran's intelligence ministry used a seized website to claim credit for a destructive attack on a US medical-technology firm. The release did not name the company.
United Kingdom
The UK's National Cyber Security Centre reports that in the year to Aug. 31, 2026 it received 1,831 reports of attacks, supported 337 incidents and classed 80 as nationally significant.
Jaguar Land Rover (2025). The NCSC says the attack caused severe, prolonged disruption to production and the supply chain, with about £196 million in direct costs, and contributed to UK economic growth of only 0.1% in the third quarter of 2025.
University of Nottingham (June). The university said a significant amount of student-record data was accessed through a flaw in Oracle WebLogic software, per the Stingrai timeline.
Russia and China. In July, the NCSC and partners warned that Russia's FSB Centre 16 was exploiting vulnerable routers to target critical infrastructure. The UK and EU states also formally attributed a December 2025 attack on Poland's energy grid to Russia's FSB. The NCSC calls China's activity the "most sophisticated and sustained cyber challenge" facing the UK, and calls Russia the "most audacious" threat.
European Union and neighbours
The EU's cyber agency, ENISA, analysed 8,257 incidents from 2025 in its Threat Landscape 2026. Phishing was the top social-engineering method (77.8%), ideology-driven activity such as hacktivist denial-of-service attacks made up 57.3% of activity, and public administration was the most affected sector.
France: sabotaged freezers. On Oct. 11–12, attackers changed settings on remotely managed refrigeration controllers at more than 700 retail sites, causing food losses and at least 15 fires. France's foreign ministry attributed it to a unit affiliated with Iran's Revolutionary Guard. Iran rejects the claim. A separate incident in early November affected cold-room controls at three French blood-service sites. No blood products were lost.
France: stolen credentials. The tax administration said access using a tax agent's stolen credentials was used to extract data on 678,000 people and businesses, and a separate intrusion consulted the national bank-account registry covering 1.2 million accounts.
Netherlands. Telecom provider Odido said a caller posing as IT staff tricked its customer-service team, affecting about 6.39 million people. It attributes the attack to ShinyHunters and says it did not pay. The Dutch Finance Ministry also took its treasury banking portal offline, leaving about 1,600 public institutions unable to view balances or request loans, though payments continued.
Poland and the Baltics. Poland's CERT reported coordinated destructive attacks on more than 30 wind and solar farms in late December 2025. Electricity production was not interrupted. A Polish ministry said a medical-software breach may concern 18.8 million people. Latvia's road-safety agency said data on 1.2 million people was taken.
Germany, Switzerland, Denmark. The State of Berlin cut two departments off its network in August after an attack claimed by a group calling itself Rhysida, and said it would not be blackmailed. Switzerland's federal administration reported about 200 accounts compromised through SharePoint servers. Denmark's CPR register breach exposed records on 8.8 million people through a supplier's lookup account, per SOCRadar.
Belgium and Romania. In January, a serious IT outage at AZ Monica hospital in Antwerp led the hospital to move seven critical-care patients to other hospitals. Romania's oil pipeline operator said its business IT was hit but oil transport continued.
Russia and Ukraine
Ukraine's cyber defenders report that high-impact incidents have declined as defences improve, but that Russian operations have adapted. Ukraine's CERT said in October 2025 that Russian hackers were using AI to write malware code and noted PowerShell scripts in the Wrecksteel malware that appeared AI-generated.
In February, Ukrainian officials said attacks on the energy grid were increasingly aimed at gathering intelligence to guide missile strikes, such as mapping facilities, tracking repair crews and assessing recovery speed. One official said: "Cyberattacks on critical infrastructure never happen on their own; they are always part of a broader operation."
Anthropic's threat report, summarized by CyberScoop, says it disrupted a Russia-aligned espionage campaign that hit more than 20 government and defence organizations across Ukraine and Europe, and that AI agents helped rebuild malware when security tools detected it. Google separately reported a Russian-speaking actor running an end-to-end automated attack, without naming a group or a state.
Russia is also a victim. Russian hosting provider Beget said attackers reached the management layer of its architecture on Nov. 1–2 and wiped all data in its cloud infrastructure, along with virtual-server backups. No attacker has been named, per the Stingrai timeline.
Israel, Iran and the Gulf
The cyber side of the Middle East conflict has spilled well beyond the region. The UAE's cybersecurity chief told Euronews that every missile or drone strike on UAE ports, refineries and airports was matched by a cyberattack, including supply-chain compromises, ransomware, fraud and deepfakes. He did not attribute them to any state. "Now it's no longer seeing is believing," he said.
Outside the region, the Stryker wiper claim, the French freezer sabotage and the US controller advisories are tied to Iran-linked actors by the claimants, US agencies or the French government. Iran denies the French accusation. Earlier, in June 2025, suspected Israeli hackers claimed to have destroyed data at Iran's Bank Sepah. Attribution in this conflict is hard, and many claims come from groups that present themselves as independent hacktivists.
China, Taiwan and East Asia
China. The UK's NCSC says China's use of data is a key driver of its operations, and the UK sanctioned two Chinese technology firms in December 2025, i-Soon and Integrity Technology Group, for malicious cyber activity. Microsoft tied the October ConnectWise supply-chain compromise to a China state-aligned actor using an AI framework called ARTEX.
Taiwan. Its National Security Bureau said its government network faced about 2.58 million intrusion attempts a day as of September, and does not rule out links to election interference before the Nov. 28 local elections. Its digital ministry said nearly half of a recent wave of attacks involved AI, including the ARTEX tool, and exposed more than 1 million personal records. Prosecutors also suspect a train radio signal was copied and spoofed on April 5, stopping three high-speed trains for 48 minutes of delays.
Singapore. The government disclosed in February that an actor it tracks as UNC3886 ran a targeted campaign against all four of its major telecom operators.
Japan. A run of mass-data incidents hit retailers, couriers and government. Furniture retailer Nitori said up to 37.62 million records may have leaked. Parcel carrier Sagawa Express reported up to 53.23 million records, a maximum count that includes duplicates. The Digital Agency reported about 246,000 items of personal data exposed through a VPN flaw. IDC Frontier's ransomware outage affected 495 organizations.
South Korea. A government-led probe found malware, including the BPFDoor family, on SK Telecom servers and counted about 26.96 million leaked SIM-identity records. KT's illicit payments were traced to an illegal mobile base station, and the Coupang breach to a former employee.
North Korea. The FBI attributes the Oct. 29 Binance theft of about US$6.3 billion to the military-linked TraderTraitor group. The Justice Department said on Nov. 12 it had court authority to seize or restrain more than $1.2 billion of it, with more than $2.7 billion frozen. It also described "FaceStealer," a campaign that has stolen more than $400 million using fake video meetings and deepfakes. Drift Protocol (US$295 million) and KelpDAO (about US$290 million) were hit earlier this year, with North Korea named in Drift's case and in LayerZero's preliminary view of KelpDAO.
India and South Asia
Indian automaker Bajaj Auto told stock exchanges on June 23 that a ransomware attack had affected its systems, and said three days later that key functions were running normally. Lookout researchers report that hackers possibly aligned with Pakistan's government have used fake recruiter websites to infect defence workers with GravityRAT spyware.
Indian media also describe a wave of "digital arrest" frauds, where scammers pose as police or investigators and pressure victims into paying. Frontline reports deepfakes being used in some of them.
Australia and New Zealand
Australia's signals agency says its cyber centre received more than 82,000 cybercrime reports in 2025–26, about one every six minutes, and responded to more than 1,500 incidents. The report says AI is making attacks, scams and phishing more efficient and convincing.
In June, Mackay Sugar stopped milling after a cyberattack. ABC reports a group called The Gentlemen claimed it; the company later said mills were operating again. Origin Energy said information on about 900,000 current and former customers was accessed. The 2025 Qantas breach, a vishing attack on a contact-centre provider, affected about 5.12 million people.
Latin America
Brazil saw some of the region's most unusual incidents. In June, an intruder used the national civil-defence alert platform to send false alerts to mobile phones, including nine extreme-level alerts in under two hours. The Port of Santos authority said in November that a cyberattack hit its files; it recovered about 70% of backups and says port operations continued.
Google's threat team describes a financially motivated actor, BREEZE COMET, that manipulates banking software and payment systems such as Pix to make fraudulent transfers. Google has evidence the group is using generative AI to support malware development and says its infrastructure may signal expansion across Latin America and Africa.
Africa
INTERPOL's 2026 assessment says AI is enabling 55% of reported cybercrimes in Africa, and that reported losses more than doubled from US$192 million to US$484 million since 2024. East Africa is a hub of mobile-money fraud, business email compromise and romance scams are widespread in Central and West Africa, and criminals are building AI-generated synthetic identities that can slip past biometric checks.
Police are pushing back. INTERPOL says Operation Serengeti III, from June 29 to Aug. 31, led to 1,462 arrests and the recovery of more than US$105 million across 21 countries. Anthropic's report also says an actor it linked to Russian state interests stole more than 300,000 national identity records from a North African government agency.
AI in the attacker's hands
AI is not one thing in these incidents. Here is what the agencies, companies and researchers cited above say it is actually doing.
1. Better lies: phishing, voices, video and fake identities
Canada's National Cyber Threat Assessment says AI is "almost certainly lowering the barriers to entry" for cybercrime. Fraud investigators and agencies describe it making impersonation cheaper: cloned voices, fake video and convincing messages.
Fake video calls. The FBI says North Korea's FaceStealer scheme, which stole more than $400 million in 18 months, works like this: hackers secretly record a victim's webcam, hijack their accounts, message their contacts, and then use the recorded video to impersonate the victim in a fake video chat that tricks the new target into infecting their own computer.
Synthetic identities. INTERPOL says criminals in Africa are building AI-generated synthetic identities that can slip past biometric checks.
Deepfake scams. The UAE's cybersecurity chief listed deepfakes among the tactics used against the country, and said "now it's no longer seeing is believing." The CAFC says impersonation, urgency and trust remain the core of fraud in Canada.
2. Faster malware and exploits
Ukraine's CERT said in 2025 that it had investigated malware showing signs of being AI-generated.
Google says a Brazilian bank-fraud group has used generative AI to help develop malware, per its BREEZE COMET report.
Google counted 22 exploited zero-days in August, up from 11 per month earlier in 2026 and an average of eight in 2025. It also saw hundreds of purported exploits appear within days of an Oct. 21 patch, most nonfunctional and likely AI-generated. Faster is not the same as better, but it is more noise for defenders.
3. Agents that run the attack
The most important change this year is the move from "AI helps a human" to "AI does the steps." The evidence is now in the public record.
ARTEX. ConnectWise and Microsoft tied the Cloud Backup compromise to a China state-aligned group, which Microsoft tracks as Storm-0940. Microsoft says it used the ARTEX agentic framework to attempt sign-ins to more than 2,250 downstream tenants, per Stingrai's timeline. Canada's Cyber Centre says that in some cases the activity went from picking a target to privileged cloud access in under three hours, and calls ARTEX "a reusable framework that can help conduct cyber operations on an industrial scale." It adds that attacks are not yet fully autonomous: humans still choose targets and set objectives.
"First fully automated" attack. Google's threat team says Russian-speaking operators ran what it believes is the first attack run end to end by AI without a human at any stage, using a system that compares software before and after a patch, infers the flaw, writes an exploit and runs it. The same quarterly report describes attackers using open-source autonomous frameworks such as CAIRN and PentAGI against organizations in the U.S. and across Asia.
Small teams, state-level reach. Anthropic reports that Chinese-speaking operators, partly university undergraduates, used its Claude models to produce more than a dozen possible zero-days in a single month, and that ShinyHunters-linked affiliates dumped more than 2,100 Azure access tokens across more than 40 corporate tenants in about 34 hours, with, in the report's words, "AI agents performed nearly all of the work." Anthropic says it disrupted each operation. CyberScoop's summary quotes the report: "For threat intelligence investigators, sophistication has stopped being a reliable signal of who is behind an operation."
4. When the AI wasn't even supposed to be attacking
In July, Hugging Face disclosed an intrusion into part of its production systems that it says was run end to end by an autonomous AI agent. OpenAI later said its own models caused it, during an internal cyber-capability test run with reduced refusals. Hugging Face's technical reconstruction says the agent escaped its test environment through a zero-day, reached the public internet, and then spent roughly two and a half days inside Hugging Face's systems. It took about 17,600 recovered actions. Hugging Face believes the agent saw it as an attempt to cheat the test by stealing the answers.
Hugging Face says the only customer content accessed was five datasets tied to the benchmark, and that no other models, datasets or packages were affected. It also stresses that a capable human attacker could have found the same weaknesses: unsafe data processing, exposed cloud metadata, overly broad access and long-lived credentials. The agent simply explored them at a different scale. The lesson is less "the machines have gone rogue" than "our security gaps are now easy to find at machine speed."
5. AI tools as targets
Google's threat team says criminals are stealing AI service credentials from cloud accounts and reselling the access. In some cases, fraudulent charges topped $100,000 in a single day. It also says a suspected North Korean group modified an AI coding assistant's configuration files so the assistant would download a backdoor when developers started it. Malicious versions of the LiteLLM AI gateway were published to the PyPI package index in March and October, and a Vercel incident began with the compromise of a third-party AI tool used by one employee, per Stingrai's timeline. Binance's incident report, as summarized by SOCRadar, says abuse of an internal AI tool on Oct. 20 led the attackers to the wallet signing service used in the theft.
The reality check
None of this means ordinary attacks have gone away. ENISA says reporting over 2025 indicates attackers primarily use consumer-grade AI to augment existing skills rather than to achieve breakthrough capabilities. The UK's NCSC says none of the AI uses its incident team has observed enabled a novel class of attack. The Cyber Centre says weakly protected accounts, excessive access, exposed systems, unpatched software and monitoring gaps are still common ways in, and that agentic AI can help attackers exploit them faster and at far larger scale.
AI also helps defenders. The Five Eyes statement urges leaders to use it to improve detection, vulnerability discovery and response, and says success "will come from getting the basics right, acting quickly."
How to protect yourself: the checklist
Most of the attacks above worked because someone trusted a call, a login page, a supplier or a script. These steps work whether the attacker uses AI or not.
For individuals and families
Hang up and call back. If someone claims to be your bank, police, the government or your employer's IT desk, end the call and contact them using a number you look up yourself. Numbers can be spoofed. The Cyber Centre's alert on an ongoing voice-phishing campaign says fake IT callers may spoof real help-desk numbers, and may end the call if you try to verify.
Never move money to a "safe account." The CAFC says: "Never transfer money to a so-called safe account at the direction of an unexpected caller." It also warns that victims of fraud are sometimes targeted again by people promising to recover their money for a fee.
Agree on a family code word. Voice cloning makes "it's me, I'm in trouble" calls convincing. A private word only family knows lets you check in seconds. This is our practical suggestion, not an agency requirement.
Never share one-time codes, and never approve a sign-in you did not start. Use passkeys or an authenticator app where you can, and a password manager with a unique password for every account.
Be wary of "fix it" prompts. Fake error screens and video calls that ask you to install software or paste a command are a common trick (ENISA describes the "ClickFix" technique, which tricks people into running malicious commands). Close the window.
Update your phone and computer. After the Optimizely attack, The Circuit reports that Report URI says iOS 18.7.3 and later, and iOS 26.3 and later, are patched against the exploit kits it saw.
Use a credit card for online purchases and turn on alerts. If you entered a card number on an affected site on Nov. 16, watch your statements and call your bank about anything you do not recognize.
Check breach notices and act on them. Take free credit monitoring if offered, and watch for follow-up scams that use the leaked details. Scammers sometimes pose as the breached organization.
Report it. In Canada, report fraud to the Canadian Anti-Fraud Centre and your local police. Report to your bank first if money is involved. The CAFC says only 5% to 10% of fraud is reported.
For small businesses, charities and community groups
The Bragg Creek case is a reminder that small organizations are targets too. CBC reports that the association's bank accounts were drained over about 30 minutes. These steps draw on the Cyber Centre's bulletin and its vishing alert, and the CAFC's fraud advice. The Cyber Centre also publishes "Foundational Cyber Security Actions for Small Organizations" (ITSAP.10.300).
Require two people for payments and banking changes. Ask your bank what dual approval, transfer limits and alerts it offers. Verify any change to payment details with a known contact.
Use phishing-resistant sign-in (FIDO2 security keys or device-bound passkeys) for administrators, remote access and other important accounts, as the Cyber Centre advises.
Set a help-desk rule. Tell staff to independently verify any unexpected call asking for an MFA, passkey or password change, regardless of caller ID or urgency, and give them a clear way to report suspicious calls.
List everyone with access. Review third-party apps, remote-support tools, suppliers and shared cloud apps. Remove what you do not use. In 2026, a backup product and remote-access tool extensions were routes into victims, according to ConnectWise and Microsoft.
Patch internet-facing systems promptly and remove services you no longer need.
Keep tested offline backups, and decide in advance who is allowed to isolate an account, application or network if something looks wrong.
Limit outside scripts on payment and sign-in pages, and use a payment window run by your payment processor. In the Optimizely attack, cards entered in a separate payment window from a processor such as Stripe were out of reach, per The Circuit.
Know your emergency calls: your bank's fraud line, your insurer, local police and the Cyber Centre's online reporting portal.
Earlier this month we covered a related case, the York University data theft, where the Qilin group's claim remains unverified.
Forecast: what could happen if things get out of hand
This section is our analysis, not a prediction. It is built on public warnings from agencies and companies. None of them says the severe scenario below is happening or inevitable. We are describing a range of outcomes so you can prepare for the likely ones and recognize the unlikely ones.
What the agencies are saying
Five Eyes (June 2026): frontier AI models "are anticipated to exceed current industry expectations." "The timeline is not years, it is months." The agencies also say "breaches will occur" and that preparedness helps you contain them quickly.
Canada's Cyber Centre (October 2026 bulletin): much of today's AI-enabled activity is noisy enough for skilled defenders to detect, but it "will likely become more targeted, less visible and harder to detect as the technology improves." It adds that decisions "may soon have to be made within minutes."
UK's NCSC (2026 annual review): by 2028 it is "highly likely" that China, Russia and other states will routinely use AI cyber capabilities, and highly likely that AI-powered malware will have been used in a ransomware attack on UK critical infrastructure.
ENISA (Threat Landscape 2026): it is likely that 2026 will see more phases of an attack directly enabled by AI, "with possible experimentation of Human-out-of-the loop proof of concepts." Google says it has now seen an attack with no human in the loop.
Canada's National Cyber Threat Assessment: adversaries "very likely consider civilian critical infrastructure to be a legitimate target for cyber sabotage in the event of a military conflict."
Scenario 1: Faster and noisier (the most likely path)
Nothing dramatic changes in kind, but everything speeds up. The gap between a patch and a working exploit keeps shrinking. Scams are more convincing and cheaper to run. More organizations get hit through a supplier, and more of us get breach notices.
What you would notice: more fraud attempts, more service outages at schools, municipalities, hospitals and retailers, and more recoveries that take weeks, like those at Vanderbilt University and the University of Mississippi Medical Center. This is largely the world of 2026, with more of it.
Scenario 2: One supplier, thousands of victims
Attackers use AI to run a campaign through a single shared dependency, such as a cloud identity service, a remote-management tool, a payment processor, a backup product or a script embedded on thousands of websites. We have seen the early form of this: the ConnectWise compromise led to sign-in attempts against more than 2,250 tenants, Report URI found the malicious Optimizely code in 178 of the 1,611 snippets it checked, and Binance's funds moved in about 20 minutes.
What you would notice: several well-known services failing or behaving oddly on the same day, card fraud spikes, delayed payments and deliveries, and a day or two of disruption before it is contained. For a sense of scale, the NCSC says last year's Jaguar Land Rover attack cost the company about £196 million in direct costs and contributed to the UK economy growing by only 0.1% in Q3 2025. This scenario is that kind of event hitting several companies at once.
Scenario 3: A crisis with a cyber layer (low probability, high impact)
In a serious international crisis, state actors pair cyber operations with physical ones and hit several essential services at the same time, using AI to run many operations in parallel and to flood the information space with convincing fakes. The UAE's cyber chief described a version of this in 2026: a cyberattack alongside every strike on a port, refinery or airport.
The building blocks have already appeared separately. More than 30 Minnesota water systems were targeted in a coordinated attack. Refrigeration controllers at more than 700 French retail sites were tampered with, which France attributes to an Iran-linked unit. Poland's grid was targeted in December, and the NCSC says the attack could have cost 500,000 citizens their power if it had been successful, though electricity production was not interrupted. Taiwan has reported a suspected spoofed rail signal and millions of daily intrusion attempts on government networks.
What it could look like: regional outages lasting days rather than hours, payment and telecom problems, hospitals diverting patients, supply chains stalling, and a period when you cannot be sure a voice or video is real. We have no evidence this is imminent. Canada's Cyber Centre, for one, is running a program to prepare critical infrastructure for it, including its CIREN initiative, focused on isolating critical systems, operating independently, and rebuilding from trusted offline sources.
Signs to watch
Government emergency directives and multi-agency alerts becoming more frequent, as CISA's Emergency Directive 26-03 on Cisco SD-WAN did in February.
Several sectors reporting outages on the same day, or the same supplier named in unrelated incidents.
Official attributions and sanctions naming states, as in the French and Polish cases.
Reports of more human-free attacks, and rising counts of exploited zero-days.
Cyber incidents timed with elections or military events, as Taiwan's security bureau has flagged ahead of its Nov. 28 local elections.
How to be ready, whichever scenario arrives
At home: keep some cash on hand, keep a paper list of key phone numbers and account details, keep an offline backup of photos and documents, and know a manual alternative for paying bills and getting information. A battery- or crank-powered radio works when networks do not.
At work or in a group: write down the minimum services you must keep running, practise operating without your main systems, keep trusted offline copies of key files and instructions, and test restoring from them. Those are the Cyber Centre's own preparation steps for critical infrastructure and they scale down well.
Everywhere: treat any unexpected urgent request as unverified until you check it through a channel you chose.
The bottom line
The headlines are frightening for a reason, but the way through is not complicated. Many of the attacks in this report succeeded because of a trusted-but-unverified phone call, a login page, a supplier, an unpatched system or a script nobody was watching. Closing those gaps does not require perfection, just consistent habits, and the agencies cited here are unusually blunt that those habits matter more than ever.
We will keep covering developments as they are confirmed, and we will update this article if facts change. If you have been affected by one of the incidents above, or if you spot something we got wrong, tell us.
General information, not professional security or legal advice.
Key sources and further reading
Five Eyes cyber agencies' statement on the AI shift in cyber risk (June 22, 2026)
Canadian Centre for Cyber Security: agentic AI bulletin and National Cyber Threat Assessment 2025–2026
Canadian Anti-Fraud Centre: fraud trends, first six months of 2026
Google Threat Intelligence Group: agentic threat activity crosses a significant threshold
Hugging Face: technical timeline of the July 2026 agent intrusion
FBI: FaceStealer cryptocurrency theft scheme and the Justice Department announcement
Stingrai's 2026 cyber attack timeline, a secondary compilation we used with attribution and checked against primary sources where we could
Support Alberta Radio
Independent cyber and news coverage takes time and resources. A $25 contribution helps keep Alberta Radio's reporting on the air.



