Hackers reached Rackspace Spot customers’ cluster secrets—how far could that access carry them?
New to this story? Here’s where it stands. Microsoft Agent Framework’s legitimate agent-framework package on PyPI received two unauthorized releases on November 17. Versions 1.26.1 and 1.26.2 installed Python startup hooks that stole environment secrets and credential files and requested cloud and CI identity tokens, without requiring an application to import the package. Both releases were removed that evening; 1.26.0 was the last clean version.
This is Lantern Cybersecurity Daily. Today: vCenter’s patch alarm, Microsoft’s package compromise, Fastify’s bypass, and a reported FBI intrusion. First, what’s at stake for Rackspace Spot customers?
This one's from BleepingComputer:
According to Rackspace, the attackers were able to act as legitimate users in some customer organizations, allowing them to create pods in Kubernetes clusters, read Kubernetes Secrets, and list nodes and volumes. Kubernetes Secrets store sensitive information such as passwords, authentication tokens, and API keys that can be used to access databases, cloud services, and other systems.
For affected Rackspace Spot customers, the exposure reaches passwords, API keys, and other credentials stored in their Kubernetes clusters. BleepingComputer reports that Rackspace says intruders used leaked credentials on November seventeenth to launch unauthorized workloads and gain access to those secrets.
Removing the intruders' workloads doesn't put those passwords back in the bottle. Rackspace says it revoked affected sessions and tokens and rotated keys, but customers still have a substantial cleanup: replace automation tokens, download fresh cluster configurations, and rotate credentials stored in cluster Secrets.
And keep the evidence straight: Rackspace warns that any Secret in an affected user's clusters could have been read; that doesn't establish that every one was stolen. A customer report alerted Rackspace on November eighteenth—customers deserve a clear account of which identities and clusters were exposed.
For a small team, start by mapping those cluster credentials to the systems they unlock. Rackspace specifically advises checking database, registry, Git, and cloud-storage logs; prioritize powerful credentials and investigate their use downstream, even after the Spot tokens are replaced.
Webforta writes:
Update 20 November 2026: CISA added CVE-2026-123890 to its Known Exploited Vulnerabilities catalog on November 20, 2026 (due date November 23, ransomware use unknown). We also added Help Net Security's details on the advisory omitting exploitation, Mandiant's evidence of pre-July knowledge, and Google's hardening and threat-hunting guidance.
For teams running VMware vCenter, this campaign already had an emergency patch. Now CISA has added CVE-2026-123890 to its exploited-vulnerability catalog, with a November 23 deadline—tomorrow—for covered federal agencies. A small engineering team can reasonably put this one ahead of the vulnerability spreadsheet.
The flaw lets attackers bypass authentication in vCenter's directory service. Webforta, citing Rapid7, reports that Mandiant and Google Threat Intelligence confirmed attacks dating back to September; Broadcom released fixes November 19. That's a substantial window to investigate alongside patching.
And the July vCenter fix doesn't cover this vulnerability. Check the installed build against Broadcom's applicable fix, including for Cloud Foundation deployments; don't let a completed July patch ticket close this job.
Also check which internal systems can reach vCenter. An internal-only deployment still deserves urgent attention here—and installing the fix doesn't establish whether someone got in during that September-to-November window.
From Rescana:
The stealer collects environment variables and credential files, and requests fresh tokens from Azure DevOps, GitHub Actions (1.26.2 only) and Azure managed identity endpoints, then sends everything to a single webhook.site address. Wiz reports that both versions were no longer available on PyPI by 16:30 EST the same day.
If your team installed Microsoft's agent-framework Python package, check the version: 1.26.1 and 1.26.2 are the malicious releases, according to GitHub's advisory. Those are two version numbers worth interrupting somebody over.
Rescana reports the credential stealer runs when Python starts in that environment—even if nobody imports the package. It collects credential files and environment variables, and requests fresh tokens from services including Azure managed identity endpoints. For affected environments, removing the package alone leaves exposed credentials to deal with.
There's also a reported entry point: Wiz suspects a stolen Microsoft release certificate and private key enabled the uploads. Suspects—we shouldn't promote that to a confirmed cause. These releases landed under Microsoft's legitimate package name, so checking the spelling wouldn't have saved you.
GitHub identifies 1.26.0 as the last clean release; the malicious versions were removed November seventeenth. Check which environments installed them and ran Python, then revoke or rotate credentials those environments could expose. Removal from the package index doesn't retrieve stolen secrets.
Here's Lawrence Abrams at BleepingComputer:
According to CNN's reporting, the FBI is investigating an apparent breach of the NCIC system, citing two people briefed on the investigation. The intrusion was reportedly discovered in recent days, although it is unclear when the breach occurred, how the attackers gained access, or whether any information was stolen. The FBI has not publicly confirmed the incident and did not respond to CNN's request for comment.
For officers checking warrants—and people whose missing-person reports or protection orders are stored there—an apparent intrusion into the FBI's National Crime Information Center carries serious stakes. BleepingComputer cites CNN's reporting, based on two people briefed on the investigation; the FBI hasn't publicly confirmed it.
And the operational questions are still open: did anyone access records, and was police access disrupted? Neither is established, so agencies shouldn't hear this report as evidence that their checks are failing.
Agreed—but agencies using this during traffic stops need a usable account of any exposure or disruption as investigators establish it. Whether information was stolen remains unclear; so do the entry point and when the intrusion happened.
And keep the earlier recruitment-platform attack separate. ShinyHunters claimed that one involved an unpatched PeopleSoft flaw; there's currently no indication connecting it to this incident. 'FBI hacked again' doesn't give investigators a shared culprit.
This one's from GitHub:
`@fastify/middie` matches path-scoped middleware against its own reading of the request path, which can differ from the path Fastify's router (find-my-way) uses to dispatch. When the two disagree, the router routes a request to a handler that a middleware mount covers while middie skips that middleware, so an unauthenticated client reaches a handler that an authentication, authorization or rate-limiting middleware was meant to protect. No special configuration is required.
For teams using Fastify's middie package: check whether your login checks or rate limits sit on path-scoped middleware. GitHub's November twentieth advisory says requests can reach the handler while skipping that protection in versions before 9.3.8.
The package and Fastify's router can read the same path differently. One example: an empty user ID can still reach a route while middie skips its middleware. The advisory says no special configuration is required.
Two path readers, one very expensive disagreement. I'd move an exposed service relying on those checks up the patch queue: 9.3.8 is the listed fix, and I'd regression-test those protected routes with the advisory's request forms.
That release aligns middleware matching with the router's path handling. The advisory describes an authentication bypass; it doesn't establish that attackers have exploited your service. Check the dependency version and where your protections run—then upgrade affected deployments.
If you follow cybersecurity, you may also enjoy The Data Center Daily, a daily briefing on AI compute, hyperscaler capex, the power grid, semiconductor supply, and energy markets reshaped by intelligence at scale. Find it wherever you listen to podcasts.
We’re watching November 23, CISA’s listed remediation deadline for vCenter CVE-2026-123890.
Links to every story are in the show notes, so take a look at the ones you want to explore further. That’s Lantern Cybersecurity Daily for today—thanks for listening, and we’ll be back tomorrow. This is a Lantern Podcast.