Nine platforms disclosed October incidents

Wiz published this post at 18:51:44 U.S. Eastern on 30 October 2026, which is 22:51:44 UTC and 06:51 Beijing time on 31 October. The article says at least nine AI developer platforms disclosed incidents during the month: Fireworks AI on 6 October, Andromeda on 9 October, Zeabur on 14 October, Lightning AI on 15 October, OpenCode on 16–17 October, Railway on 20 October, Warp on 23 October, CommandCode on 28 October, and Northflank on 30 October. Those dates are when the disclosures fell in October, not the publication time of this roundup.

The article says most of the incidents are still being analyzed, and that detailed technical information is not available. Zeabur shared a full write-up and said the company will shut down because of the incident. Railway released a preliminary report and a timeline. Fireworks, OpenCode, and Warp reported indications of what they call an agent-swarm attack. Andromeda's incident report calls it part of an industry-wide attack. OpenCode and CommandCode attributed their incidents to an attempt at distillation. The article also says no single vulnerability or dependency links these incidents. Those sentences are public characterizations, not a description of how anything was done. The post is a Wiz roundup, not a statement issued jointly by the nine companies. It says few other reports are connecting the incidents, and it names the Cloud Security Alliance as an exception, without setting out that group's findings. The post also points its own customers to advisories on the Railway incident, the Lightning AI incident, and the wider set. Those advisories are not summarized here.

On impact, Wiz limits what it will cite so far to anecdotal reports on social media. One of those reports alleges about $371,000 in cryptocurrency losses tied to the Railway incident. Another says abused access on the OpenCode side produced consumer bills in the hundreds to thousands of dollars. The article labels both as anecdote. They are not a joint audit by the nine companies, and they are not measurements Wiz claims to have made itself.

Wiz then says it observed downstream abuse of enterprise credentials and attributes that abuse to one of the vendor incidents. In both cases the key was later remediated. One case ran about thirty hours: about 800,000 Bedrock calls, with about 90 percent succeeding. The other ran about thirteen hours: about 1.7 million Bedrock calls, with more than a million succeeding. At list price, the two cases together likely ran past $100,000 of usage. The article says it cannot definitively establish the relationship between reselling organizations and either the initial compromise or the later volume of calls. The call counts and the $100,000 figure are Wiz's tally and estimate. The two tallies are not presented as a sum of every incident on the nine-name list. The article attaches them to two keys that had seen heavy use from a vendor before the unauthorized calls, and it says each key was remediated after the busy stretch. The list-price figure above $100,000 is an estimate for those two cases together, not a bill assigned to one named platform.

The disclosures on the list start on 6 October and run through 30 October. The roundup itself went out on the evening of 30 October in the Eastern time zone, before the Beijing noon window, so this edition treats it as already public results. The rest of the post describes how calls were made, which addresses were used, and which sites were involved. That is procedure, and it is not repeated here. The picture is the title card at the top of the post, not a scene from an incident.