New Delhi: Amazon has ordered a 90-day internal safety reset following a series of coding errors that had interrupted its e-commerce platform and cost it millions of orders over the past few weeks. The firm is increasing code-review restrictions following failures that revealed vulnerabilities in the software deployment procedures at the firm.
The relocation follows multiple events since the end of 2025 that influenced the operation of the online marketplace of Amazon. Documents reviewed by Business Insider indicate that at least one of the disruptions concerned the AI-assisted coding assistant Q of the company; the rest of the disruptions were associated with a wider gap in supervision and protection during the software updates.
Amazon tightens code controls after outages
The Amazon senior vice president of e-commerce services, Dave Treadwell, informed employees that an incident trend had risen since the third quarter of 2025. Some of the key failures have been experienced within the recent weeks, forcing leadership to revisit the process of approving and deploying code changes.
There were some issues due to what Treadwell termed ‘high blast radius changes’, in which updates were disseminated across systems without adequate protection. In some other instances, critical controls like two engineer approvals being made on code changes were absent or avoided.
To mitigate the problem, Amazon will add higher documentation and more levels of approval of engineers implementing changes. The company said the measures will introduce “controlled friction” to slow down risky changes in core retail systems.
AI coding tools add new risks
Amazon executives remark that the events show the increasing level of sophistication posed by generative AI coding tools. AI development tools, including Amazon’s Q, enable engineers to produce bulk code in a short period of time. Nevertheless, such an influx of code may swamp conventional review mechanisms unless protective mechanisms are revised.
Treadwell stated that the company is to unite AI-based agentic protection with more predictable, rules-based deterministic ones. It is aimed at making sure that critical infrastructure, e.g., product listing, pricing systems and order processing, is stable despite a faster pace of development of AI tools.
Millions of orders lost during march incidents
A huge failure on March 2 resulted in wrong delivery estimates throughout Amazon marketplaces. The problem has cost the company almost 120,000 cancelled orders and about 1.6 million errors on its websites as per the insider report. Investigators indicated that Amazon had a Q tool that contributed to the incident.
Only three days later, a second outage caused a 99 per cent decline in orders in North American marketplaces, causing approximately 6.3 million orders to be lost. Internal reports say that the change in production that had been implemented without the necessary documentation and authorisation was part of the disruption. The report indicated that one operator could push a change in configuration with a broad system which did not have automated validation measures.
90-day safety reset introduced
To avoid additional disruptions, Amazon is applying temporary safety regulations to approximately 335 Tier-1 systems, which are services that have a direct impact on the customers and have been hit by repeated order-altering incidents.
The policy has engineers seeking at least two reviews before the implementation of code changes and internal approval tools complying with the standards of reliability engineering at Amazon. The company is also requesting the system owners and top executives to audit production code changes in their teams.
One of the Amazon representatives added that the steps are included in the continuous work of the company on increasing the reliability of the platform and that only one of the incidents considered was directly related to an AI tool.