A single software fault in shared infrastructure can knock hundreds of companies offline at once, and as reliance on the shared cloud grows, such failures will become more frequent. That is the message from recent outages at Microsoft and JPMorgan Chase.
During November's Microsoft outage, the servers that users connect through to reach Microsoft 365 services began to fail. An automatic safeguard rerouted requests to machines that were still responding, but the extra load triggered the same software bug on them. According to the company's preliminary report, at the worst moment more than 80 percent of the servers in the affected network gateways were not responding. Email users were hit hardest, but some features of the Copilot assistant were affected too.
A few days later, a banking failure showed how far the consequences of one bug can reach. A software error at JPMorgan Chase delayed payments even for customers of other banks. Their own bank could be working perfectly well: it was enough for the other side of the transaction to be a Chase account.
According to Uptime Institute's May report, the rate of outages per individual site has been falling for the fifth year in a row, although the pace of improvement has slowed. Individual data centers may become more reliable, but a failure at one of them affects more companies if more services have come to depend on it.
Take three online shops: one uses AI to sort complaints, the second to process orders and the third to answer customers. Each may have its own software supplier. But if all three apps run on the same cloud service, a single fault can paralyze all three shops. The more a business depends on shared infrastructure, the easier it is for a failure that a few customers would once have felt to turn into a large-scale outage.
Artificial intelligence can help detect a fault in time or shorten its duration. But analysts note that during an outage AI may lose some of the data it needs to make decisions. If it is allowed to change settings on its own, a mistaken intervention can widen the failure. Microsoft showed how such a chain starts even with ordinary automatic traffic rerouting.
If the AI that sorts a shop's complaints stops working, staff can take over by hand. For that they need access to the original messages and an understanding of which requests to handle first. If the only way to reach the messages is the very app that has gone down, even a large workforce will not help. The ability for humans to intervene and restore operations is part of the AI risk management framework from the US institute NIST.
After the Microsoft outage, the Australian Computer Society, a professional body, renewed its call for a nationwide review. In its view, the government should find out which critical systems depend on a single supplier and where a suitable alternative exists.
Around Prague, based on reporting by novinky.cz