Hiddenlayer, Inc.

Tanner Burns

Kenneth Yeung

Kwesi Cappel

James Ballard

Chris Sestito

David Beveridge

Thomas Bonner

Jacob Rideout

Marta Janus

Eoin Wickens

The inputs and/or outputs of a generative artificial intelligence model are monitored to determine whether they contain or otherwise elicit undesired behavior by the model such as bypassing security measures, leaking sensitive information, or generating or consuming malicious content. This determination can be used to selectively trigger remediation processes to protect the model from malicious actions. Related apparatus, systems, techniques and articles are also described.

Generative artificial intelligence model protection using output blocklist

A machine learning model is scanned to detect actual or potential threats. The threats can be detected before execution of the machine learning model or during an isolated execution environment. The threat detection may include performing a machine learning file format check, vulnerability check, tamper check, and stenography check. The machine learning model may also be monitored in an isolated environment during an execution or runtime session. After performing a scan, the system can generate a signature based on actual, potential, or absence of detected threats.

Scanning and detecting threats in machine learning models

An analysis engine receives data characterizing a prompt for ingestion by a generative artificial intelligence (GenAI) model. The analysis engine, using a prompt injection classifier determines whether the prompt comprises or is indicative of malicious content or otherwise elicits malicious actions. Data characterizing the determination is provided to a consuming application or process. Related apparatus, systems, techniques and articles are also described.

Generative artificial intelligence model prompt injection classifier

An output of a GenAI model responsive to a prompt is received. The GenAI model is configured using one or more system prompts including one or more Easter eggs. The output is scanned to confirm whether an Easter egg is present. In cases in which at least one Easter egg is present, one or more remediation actions can be initiated to thwart an information leak by the GenAI model. Related apparatus, systems, techniques and articles are also described.

Generative AI model information leakage prevention

An analysis engine receives data characterizing a prompt for ingestion by a generative artificial intelligence (GenAI) model. The analysis engine, using a prompt injection classifier determines whether the prompt comprises or is indicative of malicious content or otherwise elicits malicious actions. The prompt injection classifier can be trained using a dataset generated by populating benign content and malicious content into a plurality of different prompt attack structures at pre-defined locations. Data characterizing the determination is provided to a consuming application or process. Related apparatus, systems, techniques and articles are also described.

GenAI prompt injection classifier training using prompt attack structures

An analysis engine receives data characterizing a prompt for ingestion by a generative artificial intelligence (GenAI) model. The analysis engine, using the received data, determines whether the prompt comprises personally identifiable information (PII) or elicits PII from the GenAI model. The analysis engine can use pattern recognition to identify PII entities in the prompt. Data characterizing the determination is provided to a consuming application or process. Related apparatus, systems, techniques and articles are also described.

Generative artificial intelligence model personally identifiable information detection and protection

An analysis engine receives data characterizing a prompt for ingestion by a generative artificial intelligence (GenAI) model. The analysis engine, using a determines using, for example, a classifier or blocklist, that the prompt comprises or is indicative of malicious content or otherwise elicits undesired model behavior. Similarly, outputs of the GenAI model can be analyzed to determine whether they comprise malicious content or cause the model to behave in an undesired manner. The output is inputted into a GenAI model along with obfuscation instructions to generate an output which is returned to the requesting user. Related apparatus, systems, techniques and articles are also described.

Generative artificial intelligence model output obfuscation

An output of a generative artificial intelligence (GenAI) model is received which is responsive to a prompt by a requestor. The output is tokenized to result in a plurality of tokens. These tokens are then used to determine that the output includes at least one string comprising personally identifiable information (PII). This determined can use pattern recognition to identify tokens and sequence of tokens indicative of PII. Thereafter, a classifier is used to assign a PII type to each string in the output comprising PII. It is then determined that at least one of the PII types in the output requires redaction which results in strings having a PII type determined to require redaction to be redacted which, in turn, results in a modified output for transmission to the requester. Related apparatus, systems, techniques and articles are also described.

Selective redaction of personally identifiable information in generative artificial intelligence model outputs

Adversarial attacks on a machine learning model are detected by receiving vectorized data input into the machine learning model along with outputs of the machine learning model responsive to the vectorized data. The vectorized data corresponds to a plurality of queries of the machine learning model by a requesting user. A confidence level is determined which characterizes a likelihood of the vectorized data being part of a malicious act directed to the machine learning model by the requesting user. Data providing the determined confidence levels can be provided to a consuming application or process. Multi-tenant architectures are also provided in which multiple machine learning models associated with different customers can be centrally monitored.

Machine learning model adversarial attack monitoring

A system detects and responds to malicious acts directed towards machine learning models. Data fed into and output by a machine learning model is collected by a sensor. The data fed into the model includes vectorization data, which is generated from raw data provided from a requester, such as for example a stream of timeseries data. The output data may include a prediction or other output generated by the machine learning model in response to receiving the vectorization data. The vectorization data and machine learning model output data are processed to determine whether the machine learning model is being subject to a malicious act (e.g., attack). The output of the processing may indicate an attack score. A response for handling the request by a requester may be selected based on the output that includes the attack score, and the response may be applied to the requestor.

Growing community of inventors

Austin, TX, United States of America

Tanner Burns