AI Cybersecurity Companies Warn
More than 24,000 internet-exposed servers are leaking authentication password hashes because of a decades-old vulnerability in their Baseboard Management Controller (BMC) interface. The exposed systems are vulnerable to CVE-2013-4786, an IPMI 2.0 authentication weakness that's been sitting there since the protocol was introduced in 2004.
Researchers at Lava, a cybersecurity and infrastructure startup, found that the sheer scale of this exposure is staggering. And it's not just a theoretical risk anymore — at least one attacker has already slapped a ransom note on an exposed management interface, demanding 0.3 BTC.
The Scale of the Exposure
Lava's team scanned the internet for publicly accessible IPMI services on UDP port 623 and found some uncomfortable numbers:
- 36,872 internet-exposed hosts with publicly accessible IPMI services
- 24,650 of those hosts expose password-derived authentication material usable for offline password-cracking attacks
- 6,240 hosts accept an empty username during authentication and are protected by weak passwords
- 2,340 instances use weak administrator passwords that match public dictionaries
The United States sits at the top of the exposure map, accounting for 39% of these vulnerable servers. That's not a rounding error — that's nearly 4 in 10 of the affected systems.
How the Flaw Actually Works
BMCs are processors baked directly into a server motherboard. They let administrators remotely manage the system completely independently of the operating system. You can power servers on or off, update firmware, mount virtual media — all the low-level stuff that matters when you're managing a data center.
The vulnerability lives in IPMI 2.0's authentication mechanism. It lets an attacker request an authentication response during the handshake, then take that response offline and crack the password using dedicated GPU rigs or similar setups. In the real world, recovered credentials often work across multiple management interfaces within the same environment. A single compromised BMC can become a pivot point to the broader management plane.
That's bad on its own. It's worse when you consider what BMC access actually enables:
- Change low-level system configurations
- Apply malicious firmware updates (see U-Boot Bootloader Under Fire: Six Critical Vulnerabilities Could Let Attackers In Through Your Firmware)
- Mount virtual media to inject malware
- Potentially pivot to other systems in the management plane
Why AI Infrastructure Is Especially Vulnerable
Here's where this gets personal for organizations running AI workloads. In environments where multiple tenants share physical hardware through virtualization or GPU partitioning, a single compromised BMC could potentially expose several customer workloads at once.
As Lava researchers put it: "A physical GPU server can support multiple tenants or workloads through virtualization, GPU partitioning, or other sharing mechanisms. In those environments, compromise of one physical server could disrupt or expose several customer workloads."
Poorly segmented AI infrastructure turns one BMC compromise into a multi-tenant disaster. And most organizations haven't segmented their management networks the way they should.
The Supermicro Pattern
One of the more striking findings from Lava's research is that a large number of the exposed BMCs are Supermicro systems protected by a 10-character uppercase password printed on the chassis label, with the username "ADMIN" in every instance.
The format theoretically provides plenty of headroom for a strong password. But its constrained structure, uppercase letters only, fixed length, predictable format, makes offline cracking practical. For comparison, researchers estimated that recovering an HPE factory password would take about 1 day per captured response on an Apple M3 system. For at least a third of the vulnerable servers, they were able to find the correct password using dictionaries and the patterns on factory stickers for default credentials.
Signs of Active Exploitation
During their research, Lava discovered an internet-exposed HPE iLO 4 login page displaying a ransom note demanding 0.3 BTC. This isn't proof of widespread exploitation, or even of successful attempts. But it does show that at least some malicious activity is underway, and someone is testing these exposed interfaces right now.
Vendor Responses
The researchers notified Supermicro in June. The company acknowledged the risk but pointed to its official guidance for administrators, which recommends rotating default BMC passwords and isolating management networks. Supermicro said it would review stronger default password policies for future hardware revisions.
Lava also notified HPE and received only a standard auto-response message. No follow-up from the vendor's security team.
What Security Teams Should Do Now
The researchers recommend four critical measures:
- Keep IPMI and Redfish off the public internet, these management interfaces should never be exposed to untrusted networks
- Rotate factory BMC passwords, default credentials are easily guessable
- Restrict access to isolated management networks, segment your infrastructure properly
- Turn off legacy IPMI authentication, modern systems should use stronger authentication methods
This vulnerability represents a classic case of "patch fatigue", a known issue from 2013 that organizations have yet to address comprehensively. As InfraTrust: Why Infrastructure Vulnerabilities Deserve a Different Patching Priority outlines, infrastructure-layer flaws often get deprioritized despite their outsized impact on security posture.
For a broader look at how exposed BMCs fit into the larger threat landscape, see Exposed Server Management Controllers: AI Cybersecurity Threats in 2026.
Security teams typically log 54% of successful attacks and alert on just 14%. The rest move through your environment unseen. If you're running AI workloads on shared hardware, you can't afford to be part of that invisible majority.
Organizations need to act now to secure their BMC configurations or risk becoming the next target in what appears to be ongoing exploitation campaigns.
This article is based on research conducted by Lava security team. The findings highlight a significant gap in infrastructure security that affects organizations worldwide.