AI Sec Watch

OpenTelemetry .NET packages have a vulnerability where parsing propagation headers (headers that track request flow across services) can allocate excessive memory, potentially causing a denial of service (DoS, where a system becomes unavailable due to resource exhaustion). The issue occurs in baggage, B3, and Jaeger processing code that allocates temporary storage before checking size limits.

OpenTelemetry's dotnet implementation has a vulnerability in how it handles gRPC responses during retries. When the server sends a `grpc-status-details-bin` trailer (extra data sent with a response), the code reads a length value from it without checking if that length is reasonable, potentially allowing an attacker to force the application to allocate massive amounts of memory and crash it (a denial of service attack, or DoS). A malicious collector or someone intercepting network traffic could exploit this.

Google's Threat Intelligence teams conducted a broad scan of the public web to find real-world examples of indirect prompt injection (IPI, where an AI system reads malicious instructions hidden in websites or documents instead of following a user's original request). The study found that most prompt injection detections on the web were actually false positives (harmless content like educational articles discussing the topic rather than actual attacks), making it difficult to identify genuine threats.

OpenTelemetry's OTLP exporter (a tool for sending telemetry data, which is information about how software is performing) reads error response bodies from servers with no limit on size, potentially causing memory exhaustion if an attacker controls the server or intercepts the connection. This could crash applications by filling up their available memory.

Pipecat's LivekitFrameSerializer contains a critical vulnerability where its deserialize() method uses pickle.loads() (a Python function that reconstructs objects from binary data) on untrusted WebSocket client data without validation. An attacker can send a malicious pickle payload to execute arbitrary code on the server, potentially compromising the entire system. This affects servers using the now-deprecated LivekitFrameSerializer, especially if exposed to external networks.

Flowise, a tool for building customized AI workflows with a drag-and-drop interface, had a security flaw in versions before 3.1.0 where a speech-generation endpoint didn't require authentication (authorization bypass, where access controls are bypassed by attackers) and could decrypt stored API keys when given a credential ID. This allowed attackers to retrieve sensitive credentials like OpenAI API keys without proper permission checks.

Flowise, a tool that lets users build custom AI workflows through a drag-and-drop interface, had a security flaw in versions before 3.1.0 where the public API endpoint (GET /api/v1/public-chatflows/:id) exposed sensitive data without filtering. The flaw revealed credential IDs, plaintext API keys (secret codes used to access other services), and password fields in the raw workflow data, making it possible for unauthorized people to see this sensitive information.

Flowise, a tool that lets users build custom AI flows through a visual interface, had a mass assignment vulnerability (a bug where user input can change database fields that shouldn't be user-controllable) in versions before 3.1.0 that allowed authenticated users to overwrite existing document storage objects and access objects from other workspaces, potentially breaking access controls (IDOR, or insecure direct object references, where an attacker can access resources by guessing their IDs).

Flowise, a tool for building customized AI language model workflows through a visual interface, had a security flaw in versions before 3.1.0 that let attackers reset any user's password without authorization. The vulnerability existed because the password reset function didn't verify that a valid reset token had been created, so attackers could submit a request with an empty or null token value (which is the default) to change a user's password if they knew the victim's email address.