When payload encryption is enabled, failure messages and stack traces that temporal-proxy is configured to keep from the upstream Temporal service can reach it in plaintext, where anyone who can read the Workflow's history at the upstream can read them. When the proxy returns a failure to a Worker, it decrypts the failure's encoded attributes and copies the message and stack trace back onto the failure, keeping the attributes. A Worker that sends that failure back through the proxy, as SDKs do when an Activity failure becomes the cause of a Workflow failure, sends the restored text with it, and the proxy forwards the text unchanged because the failure already has encoded attributes. The upstream then stores the text in plaintext beside its encrypted copy. This affects deployments that enable failure encryption in the proxy, and deployments that leave it off while the SDK encodes failure attributes itself. Payloads are not affected. temporal-proxy versions 0.9.0 through 1.2.0 are affected. The issue is fixed in 1.0.2, 1.1.2, and 1.2.1, and releases before 1.0.0 receive no fix. Upgrade to the fixed release for the deployed minor release line.
This vulnerability exists in e-Sushrut HMIS due to the exposure of internal hospital operational statistics through data structure embedded in the HTML response of the login page. An unauthenticated remote attacker could exploit this vulnerability by accessing the affected login page and extracting the returned HTML content.
Successful exploitation of this vulnerability could allow the attacker to gain unauthorized access to internal hospital operational information associated with the targeted healthcare organization.
IBM Controller Disclosure Management version 1.0.0.0 has a vulnerability where sensitive EXIF metadata from uploaded images is not properly removed before storage, potentially allowing an attacker with limited privileges to access this information. The vulnerability has a medium severity rating with a CVSS score of 4.3.
A vulnerability in the undici HTTP client for Node.js allows sensitive headers such as Authorization and Cookie to be improperly retained and sent to cross-origin redirect destinations when headers are supplied as an array of name-value tuples. This bypasses intended cross-origin header stripping, potentially exposing origin-bound credentials. The issue affects versions 7.0.0 through 7.30.1 and 8.0.0 through 8.12.3 and is fixed in versions 7.30.2 and 8.12.4.
CallBackery versions prior to 0.62.0 for Perl contain a vulnerability where the getUserConfig function improperly exposes sensitive information. Non-administrative users can access their own password hashes and administrator-only notes, which should be restricted. This flaw allows attackers with a valid session token to retrieve password hashes and potentially perform offline cracking. The issue is fixed in version 0.62.0.
A vulnerability in the follow-redirects Node.js library versions 1.16.0 up to but not including 1.16.1 allows sensitive credential headers such as Authorization, Proxy-Authorization, and Cookie to be exposed to unauthorized remote servers. This occurs because the header removal logic fails when headers are provided as an array, causing these sensitive headers to be forwarded during cross-origin redirects without warning. The issue is fixed in version 1.16.1.
### Summary
Pydantic AI's OpenTelemetry instrumentation supports `InstrumentationSettings(include_content=False)` to exclude message content — user prompts, model completions, tool call arguments and responses — from exported telemetry, so that agents can be monitored without sending sensitive content to the observability backend.
Retry prompts — the feedback messages Pydantic AI sends back to the model when its output fails validation or an output validator raises `ModelRetry` — were not covered by this setting when they were not associated with a tool call, as with structured output modes that don't use tool calls (e.g. `NativeOutput` or `PromptedOutput`) and output validators applied to text output. Their full content was recorded on model request spans even with `include_content=False`. Since validation feedback can quote invalid values from the model's response, message content that the setting was configured to withhold could reach the telemetry backend.
### Details
Retry prompts without an associated tool call were serialized into span message attributes (such as `gen_ai.input.messages` and `pydantic_ai.all_messages`) as regular text parts, and this path did not apply the `include_content` check that every other message part applies. Retry prompts tied to a tool call, and all other message content, were redacted correctly.
### Impact
This does not grant an attacker any new access to the agent or its data: the content is only visible to whoever can read the exported telemetry. The exposure matters when traces are exported to a destination whose audience is broader or less trusted than the agent's own data — a third-party observability vendor or a shared backend — which is the scenario `include_content=False` exists for. The disclosed content is limited to retry feedback: validation error details, which can quote invalid values from the model's output, or output-validator `ModelRetry` messages.
### Who Is Affected
You are affected if you enabled instrumentation with `include_content=False` and your agent can produce retry prompts outside of tool calls: it uses a structured output mode that doesn't rely on tool calls (such as `NativeOutput` or `PromptedOutput`), or output validators on text output. You are not affected if you don't set `include_content=False`, or if all of your agent's retries come from tool calls (including the default tool-based structured output mode), which were redacted correctly.
### Remediation
Upgrade to a patched version; retry prompt content now honors `include_content=False` like all other message content. If you rely on this setting, consider reviewing previously exported traces for retry prompt content that should not have been recorded.
### Workaround for Unpatched Versions
If you cannot upgrade, scrub or drop the message attributes (`gen_ai.input.messages`, `gen_ai.output.messages`, `pydantic_ai.all_messages`) in your telemetry pipeline (for example with an OpenTelemetry Collector processor), or use tool-based structured output modes, whose retry feedback honors `include_content=False`.
### Credits
Reported privately by the University of Sydney Security Research Team (Liyi Zhou, Ziyue Wang, Strick, Maurice, Chenchen Yu), and later independently discovered and reported publicly, with a fix, by @sean-kim05.
Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 0.3.4 until 1.107.6 and 2.44.0, OpenTelemetry instrumentation configured with InstrumentationSettings(include_content=False) can still export sensitive agent content through exception.message and exception.stacktrace events, error status descriptions, and model_request_parameters containing instructions or the prompted_output_template. The exposed data is available to readers of the configured telemetry backend and can include tool feedback, provider error bodies, runtime instructions, and structured-output templates even though message attributes are redacted. This issue does not grant new access to agent data, and deployments that do not use include_content=False are not affected by the setting bypass. This issue is fixed in versions 1.107.6 and 2.44.0.
Improper removal of sensitive information before storage or transfer vulnerability in Wikimedia Foundation's Mediawiki - FlaggedRevs extension through 1.46.0.
IBM PowerVM Hypervisor FW1120.00 through FW1120.01, FW1110.00 through FW1110.31, FW1060.00 through FW1060.81, and FW950.00 through FW950.H3 is affected by a vulnerability in the partition resource dump interface. An attacker with authenticated administrator-level access to the HMC or service processor can obtain a limited snapshot of partition processor state. Successful exploitation results in a confidentiality impact to the managed system.
HCL DevOps Deploy / HCL Launch contains an information disclosure vulnerability related to improper redaction of secure property values. When a deployment uses a secure property starting with certain non-ASCII characters, the redaction engine may fail to mask ASCII secure values embedded in insecure properties, potentially exposing sensitive information.
CVE-2026-73440 is a vulnerability in Arista Networks EOS affecting SNMPv3 user credentials. On affected versions, SNMPv3 local or remote user credentials may be exposed as a hashed, localized key within running and sanitized configurations. An authenticated user with access to this information could misuse it to perform unauthorized SNMP read operations or send fraudulent trap notifications. The issue was discovered internally by Arista with no known exploitation in the wild. A fix is available starting from version 4.33.0F.
A security issue in MongoDB Server's query statistics serialization on the router allows users with monitoring privileges to access unredacted search query text from other users' operations. An improper conditional check in the serialization logic causes the data redaction mechanism to be bypassed when processing search queries through the sharded cluster router. This results in sensitive query literals being stored and made accessible through the query statistics interface.
CVE-2026-53604 is a vulnerability in forgekeep's nebula-mesh prior to version 0.3.8 where the web handler fails to properly wipe the plaintext CA private key from memory after use. This improper removal of sensitive information allows an attacker with the ability to read process memory to recover the CA signing key. The issue affects only the web UI path, while the API handler correctly wipes the key. The vulnerability has been patched in version 0.3.8.
IBM UCD - IBM UrbanCode Deploy 7.2 through 7.2.3.25, and 7.3 through 7.3.2.20 and IBM UCD - IBM DevOps Deploy 8.0 through 8.0.1.15, 8.1 through 8.1.2.8, and 8.2 through 8.2.2.1 IBM DevOps Deploy / IBM UrbanCode Deploy (UCD) is susceptible to an formation disclosure vulnerability when processing redacted property values. If a deployment is configured with a secure property that starts with certain non-ASCII characters, the redaction engine may fail to mask subsequent ASCII secure values embedded inside unsecure properties. An authenticated user with permissions to view deployment request details could exploit this flaw via the UI or API to view sensitive values in plain text that should otherwise be redacted.
An access control bypass and information disclosure vulnerability exists in the base AppArmor security profile configuration of Canonical snapd. The abstraction rules located in /etc/apparmor.d/abstractions/nss-systemd (inherited via ) inadvertently permit strictly confined snap applications, which lack the privileged account-control interface, to interact directly with the io.systemd.Multiplexer and io.systemd.NameServiceSwitch UNIX domain sockets under /run/systemd/userdb/.
On systems where the systemd-userdbd service is installed and operational, the service fails to distinguish between an unconfined root user on the host system and a restricted root user running within a snap application's sandbox (such as a daemon or configuration hook). Because systemd-userdbd returns "complete" user records—including sensitive hashed user passwords from /etc/shadow—when queried by a process running as root, a compromised or malicious strictly confined snap executing code as root can successfully query the Varlink interface to retrieve all system password hashes, bypassing intended snap sandbox restrictions. This issue is mitigated by the fact that systemd-userdbd is not installed by default on standard Ubuntu deployments.
A flaw was found in the authentication configuration endpoint of the keycloak-services component, which is the core engine for Red Hat Build of Keycloak identity and access management. The issue occurs because the system fails to mask sensitive configuration values, such as reCAPTCHA secret keys, when they are requested by administrators with view-only permissions. This can lead to the exposure of third-party service credentials to unauthorized personnel or through administrative logs.
Argo CD is a declarative, GitOps continuous delivery tool for Kubernetes. From 3.2.0 until 3.2.12, 3.3.10, and 3.4.2, Argo CD ServerSideDiff can expose Kubernetes Secret values embedded in the kubectl.kubernetes.io/last-applied-configuration annotation because HideSecretData(target, live, ...) does not fully sanitize ResourceDiff.TargetState and LiveState predicted live Secret objects, allowing sensitive data, stringData, and annotations to appear in UI or CLI diffs. This issue is fixed in versions 3.2.12, 3.3.10, and 3.4.2.
In OpenStack Ironic before 37.0.1, when applying a PATCH to update fields in volume properties the user is authorized for, Ironic can return unredacted sensitive information (such as iSCSI credentials). The PATCH outcome is a security issue; the POST outcome is not a security issue.