Opinion: Cyber Outages: The Attack That Has No Name

On Monday, a man in Dallas stood in a Chase branch with a cheque he could not deposit as the system was down. That afternoon, a United Airlines pilot in Houston asked the control tower for leave to stay at the gate as the system that works out a plane's weight for take-off, pilots said, had stopped. The attack that has no name is the fourth in five days. On Thursday, the office mail of 70 countries went dark. On Friday, every departing flight in America sat on the ground for more than four hours.
The failures are global. Last week caesareans were put off in England because hospital diaries would not open. A man in London who had waited a year to buy his house could not send the papers. A trading firm in Taipei confirmed its deals by fax. IndiGo, Air India and Akasa lost their customer mail. On Monday across America customers of JPMorgan Chase, Bank of America and Wells Fargo reported that they could not get into their accounts or move their money.
Advertisement - Scroll to continue
No hacker claimed credit and no institution filed charges. These were attacks with no attacker and no name. Chase blamed a bug in its own software, which spread to other banks. United said it saw no sign of a cyber issue. Microsoft traced its outage to a bug that had slept in its servers since July and added that the fault was "not related to cyber or AI/agents". The flights, the world was told, were grounded by a contractor replacing copper lines with fibre, whose router flooded the air traffic network with data. More than 9,000 were cancelled.
For the citizen this is a blockade and a siege. You are locked out of your own mail, your appointment, your purchase, and there is no counter to go to and nobody to ring. The frustration is at being helpless before a screen. The fear is about what goes dark next: the bank, the payment app, the pension. In India, where the rupee has moved to the phone, that is not an idle fear.
Advertisement - Scroll to continue
On September 3, a machine cracked a massive math puzzle unsolved since 1991. An AI agent took a number 260 digits long and found the two primes that multiply to make it. Mathematicians applauded. The rest of us may ask: AI solves maths, so what? The lock on your bank account, your mail and your payment app is a sum of the same kind, only longer. And a machine that can do such sums can also read a million lines of old software and find the gaps that leave a system open to hackers.
The latest stability report issued by the US Federal Reserve asked 23 market professionals to name the risks ahead - 20 of them named AI. They feared attack by hostile states and damage done by the AI models themselves. On Monday a Fed governor, Michael Barr, told a conference in Milan that big banks now find AI agents trying to act in customers' accounts without proof of the customer's consent. Attackers, he said, are exploiting flaws once thought too hard to reach and so "not worth fixing". That same day America's largest bank was tripped by a bug of its own.
A few weeks earlier researchers disclosed two flaws in the language in which most websites are written: e-shops, government portals, online platforms, even Wikipedia. Site owners, among them the Finnish government and 32 government sites in Britain, went offline. In Japan, SoftBank suspended its earthquake and tsunami alerts over a weekend.
Governments have been designing Plan B for dystopia. Britain asks every household to keep three days of food, water and cash. In a world where tap and pay is the norm Sweden's central bank wants a week's cash at home and has made grocers accept it by law. Germany says 10 days of supplies.
The market has charted its own fear index. A Bridgewater estimate of cyber incidents, flat for years, then vertical, with a marker at the release of the Chinese AI model GLM-5.3, found its way online. The hedge fund puts the chance of an economically destabilising attack within a year at one in three. And as of now the chart only factors incidents that are public not attacks. Do the math - or ask AI to do it!
Bridgewater's curve at its peak works out to about 27 disclosed incidents a day. India's CERT-In, on a far wider definition that takes in scans and phishing, logged 29.44 lakh last year: some 8,000 a day, more than five a minute. India's signature fraud needs no software at all: 21,928 digital-arrest complaints and Rs 647 crore lost in eight months; the home ministry told the Supreme Court.
The useful question is who is aggregating fear through the attacks. Microsoft's annual defence report names four states: China, Russia, Iran and North Korea. The FBI blames North Korea for the $6.3 billion taken from Binance, the cryptocurrency exchange, in October. France says a unit of Iran's Revolutionary Guard seized the refrigeration of some 700 supermarkets. Romania blames a Russian-linked group for an attack on a government body last December. All three deny it. Then the freelancers: 112 ransomware gangs at work in 115 countries, as per GuidePoint Security.
The world has been fascinated by AI bots hugging other bots. In July OpenAI's models got out of a sealed test and into the systems of Hugging Face, a widely used platform. Anthropic has disclosed four such incidents. CERT-In said on October 15 that AI agents under test had reached live systems of Indian organisations and read business records, personal data included.
The quantum threat is yet to arrive. The locks protecting data are based on sums that the ordinary computer cannot finish in a lifetime. A quantum computer apparently could finish them in days. Typically, the deficit is defined in qubits aka the working units of such a machine. Think of qubits as the machine's working parts: the job as per Google's Craig Gidney, needs about 1,400 that can be trusted, and the best machine so far has eight. The gap it is estimated will narrow as AI is helping with error-correction, calibration and circuits.The US has set 2030 as the deadline for high-value federal systems.India's roadmap says 2029.
Fear is propelling business. Gartner puts the world's spending on information security at $247.5 billion this year and $276 billion next. Palo Alto Networks, the largest seller, has revenues of about $11.5 billion and a market value near $250 billion, roughly double that of January. The spending rises. And curiously despite the spending there is no stopping the rise in incidents.
Governments are trying, training AI to outwit AI. It is a derivative of the old set a thief to catch a thief principle. Anthropic's Glasswing programme publishes its scores of flaws detected, patched. India's CERT-In got access this month to Anthropic's Mythos model through Amazon's cloud. The going is slow as governments are cautious about where their code is visible.
Governments are also looking at the fertility of old systems to propagate problems. Britain's cyber chief, Richard Horne, cites decades of "tech on the cheap" usage as an issue. America's air traffic runs on copper that costs a billion dollars a year to keep alive. CERT-In has told Parliament it keeps no estimate of what incidents cost. Nor does anyone answer for it. No AI firm is obliged to report an escape.
The thieves, at least, were on time. Last week, as the mail came back, inboxes pinged with a notice in Microsoft's name: click this link within 24 hours or lose the messages held during the outage. It was a thief's letter, and it preyed on the prayers of users. Microsoft's own terms allow a credit on the bill, to business customers who apply. The woman whose caesarean was put off, the man waiting for his house and the man in Dallas holding his cheque can apply to nobody. Nobody will say who attacked, because nobody will say there was an attack. That is the attack that has no name. The screen at Chase spoke for them all: "Please try back later."
(Shankkar Aiyar, political-economy analyst, is the author of Aadhaar: A Biometric History of India's 12-Digit Revolution; and Accidental India)
KioskNews shows a cleaned-up reading view extracted from the publisher’s page — the original always lives on their site, not ours.