Skip to main content


The AI cyber threat is here

The frequency of online attacks was already increasing even before the AI revolution

The year is 2027. You wake up one morn­ing to find you can­not use email or send instant mes­sages. Your online bank account is down too. You turn on the TV to find a world in melt­down.

Hun­dreds of organ­isa­tions world­wide have been hacked, lock­ing thou­sands of employ­ees out of crit­ical sys­tems. Even­tu­ally, pub­lic trans­port breaks down and indus­trial plants explode. Gov­ern­ments hold emer­gency meet­ings to estab­lish who is respons­ible. This time the attacker is not an enemy or a ter­ror­ist cell, but a thing. The AI agents are in charge now.

This, roughly, was the vis­ion of the future con­jured up two months ago by Dario Amodei, chief exec­ut­ive of Anthropic, when he warned in a 4,000word essay that a swarm of agents could be cap­able of tak­ing over the entire inter­net within six to 12 months.

For some, parts of the proph­ecy already seem to be com­ing true. In the past week, IT sys­tems at organ­isa­tions includ­ing Microsoft, JPMor­gan, the US Fed­eral Avi­ation Admin­is­tra­tion and even the US National Secur­ity Agency have all crashed, adding to a sense of AI induced chaos. Mean­while, the cyber arm of the UK intel­li­gence ser­vices has been secur­ing tens of thou­sands of Brit­ish fridges and freez­ers.

The unusual series of out­ages and alerts has arrived at a moment when the num­ber of cyber attacks is already at record highs. So far this year, hack­ers have been accused of crimes as var­ied as the assaults on European super­mar­ket freez­ers that sparked the UK into action, com­prom­ising con­trol sys­tems in Amer­ican water util­it­ies, break­ing into the FBI and steal­ing $6.3bn from the crypto exchange Bin­ance in what was the largest rob­bery in his­tory.

For cyber secur­ity experts, though, the threat now posed by AI is not selfin­flic­ted net­work out­ages or Amodei’s pre­dic­tion of an all-power­ful autonom­ous agent — or indeed an extinc­tion level event for human­ity.

The more press­ing con­cern is that states and crim­inal groups are already using the tech­no­logy to sup­port and enhance offens­ive cap­ab­il­it­ies at a pace and scale the world is not pre­pared for.

“It’s not as bad as people want to ima­gine, but it is worse than most people real­ise,” says Robert Lee, chief exec­ut­ive of cyber secur­ity firm Dra­gos and a lead­ing author­ity on cyber attacks.

Lee is one of many cyber pro­fes­sion­als who say the focus by AI lead­ers on the­or­et­ical worst-case scen­arios is dis­tract­ing people from the very real danger that AI-assisted cyber crime and cyber war­fare already pose.

“Is defence doable? Yes,” he says. “Can we win? Yes. Are we going to? No.”

Extor­tion, phish­ing and self-inflic­ted wounds

The fre­quency of cyber attacks was already rising fast even before the AI revolu­tion. Charl van der Walt, global head of secur­ity research at Orange Cyber­de­fense, part of the French tele­coms group, says the cyber extor­tion incid­ents tracked by his team — attacks where hack­ers steal data for sale or ransom — have mul­ti­plied over the past six years from roughly 1,500 in 2020 to 7,100 in 2025. Over that time, he adds, the num­ber of act­ive cyber extor­tion groups has risen from 33 to 114.

The rapid growth has been driven by factors includ­ing increas­ing digit­al­isa­tion, the pro­fes­sion­al­isa­tion of cyber crime and the rise of crypto­cur­ren­cies that have made it easier for hack­ers to get paid. This year, van der Walt has already tracked 7,500 such hacks in the nine months to Septem­ber. While some of those attacks may have used AI tools in some way, none were auto­mated hacks run by AI, accord­ing to Orange’s research. “AI tech­no­logy is not driv­ing the secur­ity prob­lem; it is reveal­ing the secur­ity prob­lems,” he says.

In August, Amer­ican research­ers used OpenAI and Google mod­els to identify vul­ner­ab­il­it­ies in PHP, a pro­gram­ming lan­guage behind 70 per cent of all web­sites, that had been undetec­ted for more than a dec­ade.

Organ­isa­tions are now identi­fy­ing soft­ware bugs at such a rate that IT sys­tems can­not keep up with the pace of the neces­sary fixes.

US offi­cials have said the rush to update soft­ware was behind the recent out­ages that blocked access to Microsoft email in 70 coun­tries, groun­ded all air traffic in the coun­try and even dis­rup­ted a clas­si­fied internal net­work at the NSA, the intel­li­gence agency that mon­it­ors and inter­cepts elec­tronic com­mu­nic­a­tions. “Every day brings another reminder that AI is mov­ing faster than our abil­ity to man­age its con­sequences,” Mark Warner, vice-chair of the Sen­ate Select Com­mit­tee on Intel­li­gence, said of the recent out­ages.

US Pres­id­ent Don­ald Trump was more san­guine. “Tell them to relax, everything’s good,” he told Fox News on Wed­nes­day as he sought to assure Amer­ic­ans that the US was not under attack. “They’re put­ting in new equip­ment, and things hap­pen when you put in new equip­ment.”

Such assur­ances did little to quell con­spir­acy the­or­ies that the out­ages were caused by co-ordin­ated cyberter­ror­ism — or con­cern about the vul­ner­ab­il­ity of vital infra­struc­ture. By Thursday, Trump had sacked US national cyber dir­ector Sean Cairncross.

For now, says Mar­cus Hutchins, a Brit­ish cyber secur­ity researcher who has first-hand exper­i­ence of cyber crime, hack­ers are mostly using AI tools such as Chat­GPT to develop and execute phish­ing attacks through which people are tricked into hand­ing over their login cre­den­tials.

The excep­tion, Hutchins acknow­ledges, is “top tier” state threat act­ors, such as China, Rus­sia and the US, who he sus­pects are “stock­pil­ing” the most advanced zero-day exploits for future deploy­ment: “A lot of their work is around build­ing cap­ab­il­it­ies for the future, so we don’t really see what they have, only what they use.”

Mat­ters of life and death

The biggest shift so far in the bal­ance of power between hack­ers and defend­ers, experts say, con­cerns oper­a­tional tech­no­logy, known as OT.

Most hacks tar­get inform­a­tion tech­no­logy sys­tems with the aim of steal­ing money or data. OT attacks tar­get indus­trial equip­ment such as power plants and factor­ies with the object­ive of lock­ing out oper­at­ors to demand a ransom or cause dam­age.

His­tor­ic­ally, such attacks have required highly spe­cial­ised know­ledge and required months of expens­ive plan­ning. Per­haps the most fam­ous OT attack of all time, the Stuxnet mal­ware assault on Iran’s nuc­lear pro­gramme by the US and Israel in 2009, took years of pre­par­a­tion and involved build­ing a phys­ical rep­lica of Iran’s cent­ri­fuge sys­tems to test the mal­ware before deploy­ing it.

Hack­ing groups can now use large lan­guage mod­els to read the manu­als of indus­trial equip­ment and con­duct research on past acci­dents, says Lee at Dra­gos. OT is often poorly defen­ded, pre­cisely because it was tech­nic­ally com­plic­ated to attack, he notes.

“When Dario [Amodei] comes out and says, ‘Hey, I think people are going to die’, I agree with him, but for dif­fer­ent reas­ons,” Lee adds. The risk will come from human adversar­ies using AI mod­els to sharpen their attacks on crit­ical infra­struc­ture, he says. “That inter­sec­tion is dan­ger­ous.”

Lee notes that soph­ist­ic­ated offens­ive cyber cap­ab­il­it­ies have moved from intel­li­gence agen­cies in the 2010s to the mil­it­ary in the 2020s to a mot­ley mix of prox­ies, gov­ern­ment con­tract­ors and crim­inal groups today. Where intel­li­gence agen­cies tend to use such tools spar­ingly, pre­fer­ring to con­duct deni­able, cov­ert oper­a­tions, the new­est act­ors lack the same restraint. “I think it is a very safe assess­ment to say we are going to see loss of human life as a res­ult of OT cyber attacks,” he says.

When the front line is a freezer

Many west­ern coun­tries are already in con­flict, either offi­cially or unof­fi­cially, with major cyber powers. Since the US and Israel began air strikes on Iran in Feb­ru­ary, Ira­nian hack­ers have staged offens­ive cyber oper­a­tions on US domestic tar­gets, includ­ing attacks on dozens of water com­pan­ies. Last week, a cyber unit of Iran’s Islamic Revolu­tion­ary Guard Corps was blamed for Octo­ber’s unpre­ced­en­ted cyber attacks in France, which turned up the tem­per­at­ure on hun­dreds of super­mar­ket freez­ers to as high as 50C, des­troy­ing mil­lions of euros of food. Sim­ilar attacks hit super­mar­kets in 10 other coun­tries across Europe this month, prompt­ing the NCSC’s dis­creet nation­wide effort in recent weeks to safe­guard Brit­ish refri­ger­a­tion.

“I don’t think there’s going to be some kind of con­scious use by a state of larges­cale AI attacks out­side of con­flict, as the rules of engage­ment still apply,” says a west­ern intel­li­gence offi­cial says. “Where you have greater risk is that all our under­stand­ing of the cap­ab­il­ity is still nas­cent. Every­body’s still learn­ing, so there is more like­li­hood of unin­ten­ded con­sequences.”

European offi­cials are increas­ingly open about Rus­sia’s use of so-called hybrid war­fare, includ­ing cyber attacks, against EU mem­ber states to exert pres­sure related to its war against Ukraine. Iran and Rus­sia’s strategy is to “under­mine the con­fid­ence and the cohe­sion of the soci­ety that you’re in con­flict with”, says Orange’s van der Walt.

China has a dif­fer­ent approach. West­ern offi­cials say Beijing is more inter­ested in cap­ab­il­it­ies that enable espi­on­age or serve as a deterrent against for­eign attacks on its own people and infra­struc­ture.

In a far-reach­ing cyber cam­paign exposed in 2023, US author­it­ies found Chinese mal­ware embed­ded in Amer­ican energy, water, pipeline and trans­port­a­tion sys­tems in appar­ent pre­par­a­tion to dis­rupt, if needed, the US eco­nomy and mil­it­ary oper­a­tions. The fol­low­ing year, US agen­cies made an even more shock­ing admis­sion. Chinese hack­ers had gained and main­tained access to US tele­com­mu­nic­a­tions net­works for sev­eral years, giv­ing the coun­try’s intel­li­gence poten­tial access to the unen­cryp­ted phone com­mu­nic­a­tions of almost any­one in Amer­ica.

For Anne Neuber­ger, a former US deputy national secur­ity adviser who once led intel­li­gence oper­a­tions for the NSA, China’s con­trol of its inter­net and eco­nomy gives it a nat­ural advant­age in cyber war­fare:. “The ten­sion between an author­it­arian coun­try’s cap­ab­il­it­ies in terms of being able to mon­itor and defend and a demo­cracy’s [abil­ity to do so] become[s] even sharper in the age of AI,” she says, not­ing that the Chinese gov­ern­ment has a dir­ect role in pro­tect­ing crit­ical infra­struc­ture.

She is less con­cerned about the vul­ner­ab­il­ity of big west­ern banks with large cyber secur­ity budgets than about infra­struc­ture such as hos­pit­als, pipelines and power sys­tems, which were his­tor­ic­ally expens­ive to secure and have never been prop­erly pro­tec­ted.

These are, she adds, “sys­tems that were built over the last 50 years, well before soph­ist­ic­ated offens­ive cyber cap­ab­il­it­ies”.

Cyber secur­ity com­pan­ies are already all rolling out AI-powered defence tech­no­logy. Some help secur­ity teams to identify and ana­lyse threats. Oth­ers use agen­tic AI sys­tems to respond to and neut­ral­ise attacks. Another pos­sib­il­ity, Neuber­ger says, is to build digital twins of real-world crit­ical infra­struc­ture on which the latest fron­tier mod­els can be tested. As the cyber wars escal­ate, she main­tains there is no time to waste in using the new tech­no­logy to pro­tect, as well as attack, the infra­struc­ture that has become the new battle­ground: “AI abso­lutely helps attack­ers . . . but it’s a game-changer on defence.”

Comments

Popular posts from this blog