Arda Kucukkaya/Anadolu via Getty Images
Are AI tools reliable enough to be used in a business setting? If so, should they be given “autonomy” to take decisions? These questions are being raised after at least two internet outages in Amazon’s cloud division were reportedly caused by the fault of AI agents. new reporting From financial Times.
In one incident in December, Amazon Web Services engineers allowed its in-house Kiro “agent” coding tool to make changes, causing a 13-hour disruption, according to four sources familiar with the matter. Unfortunately, the AI ​​had decided to “remove and recreate the environment,” sources said.
Amazon employees claimed that this is not the first service disruption related to the AI ​​tool.
“We have already seen production disruptions at least twice (in the last few months),” a senior AWS employee said. foot. “Engineers let the AI ​​(agent) solve the problem without any intervention. The interruptions were small but completely predictable.”
AWS launched its in-house coding assistant, Kiro, in July. The company describes the tool as an “autonomous” agent that can help take projects “from concept to production.” Another AI coding assistant developed by Amazon, described as AI Assistant, was involved in the earlier outage.
Employees said AI tools were treated as an extension of an operator and given operator-level permissions. In both outages, going against normal protocol, engineers did not require anyone else’s approval before finalizing changes.
in a statement to footAmazon claimed that the outage was an “extremely limited incident” that only affected one service in parts of China. Furthermore, it was a “coincidence that AI tools were involved” and “the same problem could have occurred with any developer tool or manual action,” it said.
It also claimed that its Kiro AI “requests authorization before taking any action”, but the engineer involved in the December outage had more permissions than usual, calling it a “user access control issue, not an AI autonomy issue”.
Amazon stressed, “In both cases, it was user error, not AI error.”
The company also claimed it had not seen evidence that mistakes were more common with AI tools. To which we respond: Is Amazon living under a rock? While AI and its penetration into business applications is still in its infancy, there is no shortage of evidence showing that the tools are prone to malfunction. Their tendency to produce hallucinations, or instances in which they fabricate facts, is well documented. So are their weak guardrails. He pointed out that even some Amazon employees are reluctant to use AI tools because of the risk of error foot.
Experienced programmers are finding that AI coding assistants constantly spit out bad code, with multiple studies showing that repeatedly double- and triple-checking for suspicious output actually slows down software engineers, even if the AI ​​is, on a surface level, producing faster code. The rise of “vibe coding” with AI has led to many mistakes in which an agent AI makes decisions that were not intended by its owners.
Of course, this wouldn’t be a big deal if tech companies weren’t using AI tools they claim will supercharge productivity in their own operations, and they’re willing to get high on their supply. Both Microsoft and Google claim that more than a quarter of their code is now written with AI. Engineer at Anthropic and OpenAI is suggested Almost 100 percent of their code is written AI.
It’s also ridiculous for Amazon to dismiss the outage as simple user error rather than AI. AI was used to generate the code. And Amazon, as well as its competitors, are constantly telling their employees and customers that they should rely more on the tool. employees told foot The company set a target for 80 percent of developers to use AI for coding tasks at least once a week. This is nothing less than a mandate to use AI. But if the AI ​​goes awry, it will be the employee’s fault, never the AI’s — or, for that matter, the boss pushing it.
As far as we know, the newly revealed AI glitches are unrelated to the massive AWS outage that took out half the internet last October. But in light of these revelations, as well as the company’s increasing reliance on AI tools, you have to wonder if technology was somehow involved in that disaster.
More on AI: Look, OpenAI: Tangle admitted AI ads were a mistake, now getting rid of them
