Amazon’s massive cloud computing arm has reportedly experienced at least two outages caused by its own artificial intelligence tools, raising questions about the company’s embrace of AI as it lays off human employees.
Amazon Web Services (AWS) operations were disrupted for 13 hours in December after an AI agent autonomously chose to “delete and then recreate” part of its environment, the Financial Times reported.
AWS, which provides much of the Internet’s critical infrastructure, has experienced several outages in the past year.
An incident in October brought dozens of sites down for hours and sparked debate over the centralization of online services on infrastructure owned by a handful of giant companies. AWS has won 189 UK government contracts worth £1.7bn since 2016, the Guardian reported in October.
The company said the disruptions caused by AI were minor incidents and only one affected customer-facing services.
Amazon confirmed plans to cut 16,000 jobs in January, after laying off 14,000 staff last October. In January, its chief executive, Andy Jassy, reported that the cuts were about company culture, not about replacing workers with AI.
However, Jassi has previously said that efficiency gains from AI will “reduce” Amazon’s workforce in the coming years, and allow AI agents to “focus less on thinking strategically about how to improve customer experiences.”
In a statement to the FT, Amazon said it was “coincidental” that AI tools were involved in the outages and that there was no evidence that such technology led to more errors than human engineers. “In both cases, it was user error, not AI error,” it said.
Many experts have expressed doubts about this estimate. A security researcher, Jamieson O’Reilly, said: “While engineering errors are rarely caused by traditional tools and humans, the difference between accidents involving AI and those ‘without AI’ is that a human usually has to manually type a set of instructions and they have plenty of time to realize their own error while doing so.
O’Reilly said AI agents are often deployed in constrained environments and for specific tasks, and cannot understand broad ramifications like restarting a system or deleting a database — which may have led to the error at Amazon.
“They don’t have full visibility into the context they’re running in, how your customers are affected or how much downtime at 2 a.m. on a Tuesday is costing them,” he said.
“You have to constantly remind these tools in context — ‘Hey, this is serious, don’t upload it.’
Last year, an AI agent built by tech company Replit to create an app scoured the company’s entire database, generated reports and then lied about its actions.
Michał Woźniak, a cyber security expert, said that it is almost impossible to completely prevent Amazon’s internal AI agents from making future errors, because AI systems make unexpected choices and are very complex.
“Amazon never misses an opportunity to refer to “AI” when it’s useful to them — in the case of mass layoffs designed to replace engineers with AI. But when the slap generator is interrupted, suddenly it’s just a ‘coincidence,'” he added.
Amazon has been contacted for comment.
