Independent researcher Ben Zimmermann scanned 7.25 million public Gitee repositories and found 26,400 cloud and AI credentials that still worked. They included thousands of Alibaba Cloud, DeepSeek, Zhipu BigModel and OpenAI keys, and he could not confirm that any had been revoked. More info here: 26,400 Valid Cloud and AI Credentials Found Exposed in Public Gitee Repositories
Separately, exploitation of a critical Cisco Catalyst SD-WAN flaw (CVE-2026-20127) has been tied to Volt Typhoon, with Lumen saying the group was exploiting the flaw by late 2025 and kept at it after Cisco disclosed it in February. More info here: The CISA adds eight Cisco SD-WAN flaws to KEV and gives organizations four days to fix them | The IT Nerd
Justin Beals, CEO & Founder, Strike Graph Said This:
On the exposed Gitee credentials:
“Zimmermann tested every key against the service that issued it, so 26,400 is a count of credentials that actually worked, and that makes this research more useful than most leak reports. He sent lists to the vendors with the most exposed keys, but nobody contacted the developers, and he couldn’t confirm that a single key had been revoked. If you ever pushed a key to a public Gitee repository, assume someone else still has it.
Nearly half of these credentials come from 2025, which matches how fast teams are building on AI right now. A developer grabs a DeepSeek or OpenAI key to get a prototype running, commits it, and moves on. Deleting it later doesn’t help, because the commit history keeps it and only the provider can revoke it. One of the exposed keys was an organization-admin credential for an enterprise AI platform. The two scanners agreed on only 1,001 of the keys, and each one caught about half the total. A clean result from one tool only tells you what that tool knows how to look for.
Every organization building with AI should inventory its machine credentials and give each one an owner, an expiration date, the narrowest scope that works, and a spend limit. Run more than one scanner, and revoke exposed keys at the provider instead of scrubbing the repo. As more agents move into production on keys like these, more companies will find their leaks in a cloud bill before they find them in their logs.”
On the Cisco SD-WAN exploitation tied to Volt Typhoon:
“Volt Typhoon keeps going after the network gear that sits between everything else, and SD-WAN fits that pattern. When an actor controls the device routing your traffic, they gain a view of everything moving across your network, and they blend in as part of the infrastructure. We saw the same playbook with Salt Typhoon inside the telecom carriers, and we are still working to get those actors out.
Most security programs still treat routers and edge appliances as plumbing. They get a configuration review once a year and a patch whenever the vendor ships one, while nation-state groups find the flaws first. For defense contractors and critical infrastructure operators, this goes beyond a security gap. If you can’t verify the integrity of the devices carrying your controlled information, you can’t honestly affirm that your controls work.
Network devices need to be first-class assets, with continuous configuration validation, log integrity checks, and a plan for the window before a patch exists. Expect more of this, because the edge keeps paying off for these groups.”
This is pretty bad and need urgent attention now. As in right now.
Related
This entry was posted on October 29, 2026 at 4:12 pm and is filed under Commentary with tags Strike Graph. You can follow any responses to this entry through the RSS 2.0 feed.
You can leave a response, or trackback from your own site.
26,400 working cloud and AI credentials exposed on Gitee & Cisco SD-WAN exploitation tied to Volt Typhoon
Independent researcher Ben Zimmermann scanned 7.25 million public Gitee repositories and found 26,400 cloud and AI credentials that still worked. They included thousands of Alibaba Cloud, DeepSeek, Zhipu BigModel and OpenAI keys, and he could not confirm that any had been revoked. More info here: 26,400 Valid Cloud and AI Credentials Found Exposed in Public Gitee Repositories
Separately, exploitation of a critical Cisco Catalyst SD-WAN flaw (CVE-2026-20127) has been tied to Volt Typhoon, with Lumen saying the group was exploiting the flaw by late 2025 and kept at it after Cisco disclosed it in February. More info here: The CISA adds eight Cisco SD-WAN flaws to KEV and gives organizations four days to fix them | The IT Nerd
Justin Beals, CEO & Founder, Strike Graph Said This:
On the exposed Gitee credentials:
“Zimmermann tested every key against the service that issued it, so 26,400 is a count of credentials that actually worked, and that makes this research more useful than most leak reports. He sent lists to the vendors with the most exposed keys, but nobody contacted the developers, and he couldn’t confirm that a single key had been revoked. If you ever pushed a key to a public Gitee repository, assume someone else still has it.
Nearly half of these credentials come from 2025, which matches how fast teams are building on AI right now. A developer grabs a DeepSeek or OpenAI key to get a prototype running, commits it, and moves on. Deleting it later doesn’t help, because the commit history keeps it and only the provider can revoke it. One of the exposed keys was an organization-admin credential for an enterprise AI platform. The two scanners agreed on only 1,001 of the keys, and each one caught about half the total. A clean result from one tool only tells you what that tool knows how to look for.
Every organization building with AI should inventory its machine credentials and give each one an owner, an expiration date, the narrowest scope that works, and a spend limit. Run more than one scanner, and revoke exposed keys at the provider instead of scrubbing the repo. As more agents move into production on keys like these, more companies will find their leaks in a cloud bill before they find them in their logs.”
On the Cisco SD-WAN exploitation tied to Volt Typhoon:
“Volt Typhoon keeps going after the network gear that sits between everything else, and SD-WAN fits that pattern. When an actor controls the device routing your traffic, they gain a view of everything moving across your network, and they blend in as part of the infrastructure. We saw the same playbook with Salt Typhoon inside the telecom carriers, and we are still working to get those actors out.
Most security programs still treat routers and edge appliances as plumbing. They get a configuration review once a year and a patch whenever the vendor ships one, while nation-state groups find the flaws first. For defense contractors and critical infrastructure operators, this goes beyond a security gap. If you can’t verify the integrity of the devices carrying your controlled information, you can’t honestly affirm that your controls work.
Network devices need to be first-class assets, with continuous configuration validation, log integrity checks, and a plan for the window before a patch exists. Expect more of this, because the edge keeps paying off for these groups.”
This is pretty bad and need urgent attention now. As in right now.
Share this:
Like this:
Related
This entry was posted on October 29, 2026 at 4:12 pm and is filed under Commentary with tags Strike Graph. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.