The AWS team has pushed a non-denial denial dressed up as a correction. I guess it’s time for this disinformation historian to break down what’s happening:

The Amazon post claims to “correct” an FT report while actually it confirms the core story. Burying it in minimization doesn’t actually contradict it. They admit an AI-related tool (Kiro) caused a production service disruption, then spend every sentence trying to make you forget they just said that.

For the record, an AI agent was asked to fix a bug in Cost Explorer, autonomously decided the right move was to delete the environment and rebuild it, and caused a 13-hour outage in mainland China.

The AWS rhetorical moves to avoid admitting this are so sloppy, perhaps they were written by AI:

Blame displacement: they say “user error—specifically misconfigured access controls”. Yeah, ok, this reframes an AI agent autonomously misconfiguring production infrastructure as equivalent to a human typo. The whole point of the FT story is that AI coding agents can make changes they shouldn’t. Calling the agent action a “user error” is like calling a car crash “road error.”

Wait, no, it’s worse.

FT sources said employees complained about being forced onto Kiro. So the “user error” framing is even more absurd given management was pressuring engineers to use this tool aggressively while simultaneously blaming them when it breaks things. Management demanded rapid adoption, few or no guardrails, and the result seems predictable.

Amazon set an 80% weekly usage target for Kiro and actively tracked adoption rates. They got what they asked for, an outage caused by Kiro.

Scope minimization: “extremely limited,” “single service,” “one of our 39 Geographic Regions,” “did not impact compute, storage, database”… the list goes on and on like they’re not even convinced themselves yet. Look at everything that didn’t break rather than explaining what did break and how, is an annoying rhetorical move.

The big lie: “We implemented numerous safeguards to prevent this from happening again—not because the event had a big impact (it didn’t), but because we insist on learning.” You don’t add mandatory peer review for production access over nothing. That’s a significant operational control change. Denial, denial. That tells you the existing guardrails failed.

Amazon designed a system with no guardrails, mandated aggressive adoption, then called the inevitable result “user error.” That’s just wrong. The FT reported that AI tools were treated as extensions of the operator and given the same permissions, such that operator-level access with no mandatory peer review was never a “misconfigured” anything.

The rough landing: “The Financial Times’ claim that a second event impacted AWS is entirely false.” This is the only categorical denial in the entire piece, which means everything else is carefully worded admission. Simple math. Amazon is really screwing themselves at this point. Why? GeekWire got a clarification from AWS that the second event “did not take place within the AWS business, but elsewhere within Amazon.” So the categorical denial of “entirely false” really are just weasel words about what’s technically within AWS. The second tool was Amazon Q Developer, so it’s still Amazon, still AI, and now shows a pattern.

Zero technical detail: Honestly, this is my biggest complaint. AWS used to care about engineering. For a company that once upon a time published detailed post-incident reports, there’s nothing here. No timeline, no root cause analysis, no explanation of what Kiro actually did or what permissions it had. The COE process they brag about produced what? This press release?

The whole thing reads like legal had an intern who used a chatbot prompt that said “technically don’t lie while giving the impression nothing happened”.