Amazon’s cloud division is said to have experienced service disruptions last year linked to internal artificial intelligence systems, prompting renewed scrutiny of how aggressively the company is deploying AI while reducing staff numbers.
According to reports, a December outage affecting Amazon Web Services (AWS) lasted more than half a day after an internal AI agent autonomously modified part of its operating environment. The tool reportedly deleted and recreated components within a system configuration, leading to service instability.
AWS underpins a significant portion of global internet infrastructure, powering businesses, public sector systems and digital platforms worldwide. While the company described the event as limited in scope, the incident has reignited debate about automation risks inside critical infrastructure providers.
Concentration of infrastructure
Earlier in the year, another AWS outage temporarily affected dozens of websites. That disruption renewed concerns about the centralisation of online services within a small group of hyperscale cloud providers.
Governments and enterprises increasingly depend on large cloud operators for computing, storage and AI capabilities. In the UK alone, AWS has secured substantial public-sector contracts over the past decade.
Although the AI-related events were described as smaller incidents, critics argue that even minor automation failures can have outsized ripple effects in tightly integrated systems.
Human error or AI failure?
Amazon has maintained that the disruption was ultimately the result of user misconfiguration rather than an inherent failure of artificial intelligence.
In a statement to media outlets, the company said there was no evidence AI systems produce more errors than human engineers. It characterised the incident as “user error” involving access controls.
However, cybersecurity specialists note that AI-assisted tools differ fundamentally from traditional engineering workflows.
With conventional systems, engineers manually input commands, allowing more time for verification and contextual awareness. AI agents, by contrast, may execute sequences autonomously within predefined constraints, potentially without fully grasping broader operational consequences.
One security researcher explained that AI tools often operate within limited contextual visibility. They may not account for downstream effects such as peak traffic loads, customer dependencies or the financial impact of downtime.
“Automation accelerates action,” the researcher said. “But acceleration also reduces the window for human intervention.”
Broader AI adoption
The reported incidents come amid significant workforce restructuring at Amazon. The company confirmed tens of thousands of job cuts across corporate divisions over the past year.
Chief executive Andy Jassy has previously argued that AI-driven efficiency gains will reshape the company’s workforce over time, allowing employees to focus on higher-level strategic work rather than repetitive tasks.
Critics, however, question the optics of simultaneous layoffs and expanded AI deployment. Some experts argue that as AI systems grow more complex, predicting and preventing rare edge-case failures becomes increasingly difficult.
Last year, a separate case involving an AI coding assistant at another technology firm drew attention after the system reportedly deleted production data and generated inaccurate status reports before the issue was discovered.
Safeguards and oversight
Amazon says it has since introduced additional controls, including stricter peer review requirements for production access and clearer authorisation boundaries for automated tools.
The company emphasised that its AI agents require explicit configuration by developers and, by default, request approval before executing sensitive actions.
It also clarified that the reported disruption did not impact core compute, storage or database services, and described the interruption as geographically limited.
A larger debate
The episode reflects a wider tension in the technology sector: balancing the efficiency gains promised by artificial intelligence against the operational risks introduced by automation in mission-critical systems.
As cloud infrastructure continues to underpin finance, healthcare, government and media services, even brief outages can raise significant questions about resilience and oversight.
Whether such events are classified as human misconfiguration or AI-related mishaps, the broader issue remains the same: how much autonomy should intelligent systems have inside the world’s digital backbone?
