Method and apparatus for semantic data leak prevention

Abstract

For semantic data leak prevention, one or more semantic interpretations of an input text are obtained with a generative pre-trained transformer, GPT, configured for interpreting the input text with reference to a target context, and it is determined whether the input text comprises a hidden data leakage by checking the semantic interpretations of the input text with a data leak prevention, DLP, system configured for detecting a data leakage based on a collection of words and phrases of interest. As a result, enabled is a prevention of hidden data leaks when a sensitive information is hidden within a text, for example, by using advanced language techniques.

Description

[0001] METHOD AND APPARATUS FOR SEMANTIC DATA LEAK PREVENTION

[0002] TECHNICAL FIELD

[0003] The present disclosure relates generally to the field of data security and more specifically, to a method and an apparatus for semantic data leak prevention.

[0004] BACKGROUND

[0005] In today's digitally interconnected world, the prevention of sensitive data has emerged as a paramount concern for both large enterprises and small businesses. A data leak prevention, DLP, system serves as a sentinel, which tirelessly scans and safeguards outgoing data traffic in order to detect sensitive information, such as trade secrets, proprietary intellectual property, technical specifications of products, financial records, personal information about employees and customers, and the like. Moreover, the primary purpose of the DLP system is to identify and prevent the unauthorized transmission of sensitive information from an organization's network to external destinations.

[0006] Currently, existing DLP systems utilize text-matching techniques that can be tricked by attackers, such as by using advanced literature techniques, for example, re-wording, allegory, metaphor, and the like that can be used to hide sensitive information within the seemingly neutral text. However, such attempts to protect the data leakage fail due to many reasons, such as due to inaccurate and inefficient text detection, using text shuffling or text cropping techniques in formal documents due to which the exact meaning of the data is altered, and the like. For example, a trojan can successfully bypass the DLP checks and leak sensitive information to a third party by using semantic re-encoding. As a result, there exists a technical problem of how to detect the semantic leakage to ensure overall data security and data confidentiality.

[0007] Therefore, in light of the foregoing discussion, there exists a need to overcome the aforementioned drawbacks associated with the conventional method and an apparatus of semantic data leak prevention. SUMMARY

[0008] The present disclosure provides a method and an apparatus of semantic data leak prevention. The present disclosure provides a solution to the existing problem of how to detect semantic leakage to ensure overall data security and data confidentiality. An objective of the present disclosure is to provide a solution that overcomes at least partially the problems encountered in the prior art and provides an improved method and an improved apparatus to provide semantic data leak prevention.

[0009] One or more objectives of the present disclosure are achieved by the solutions provided in the enclosed independent claims. Advantageous implementations of the present disclosure are further defined in the dependent claims.

[0010] In one aspect, the present disclosure provides a method of semantic data leak prevention. The method includes obtaining one or more semantic interpretations of an input text with a generative pre-trained transformer, GPT, configured for interpreting the input text with reference to a target context, determining whether the input text comprises a hidden data leakage by checking the semantic interpretations of the input text with a data leak prevention, DLP, system configured for detecting a data leakage based on a collection of words and phrases of interest.

[0011] Advantageously, the method of semantic data leak prevention is used for preventing data leaks that occur when any sensitive information is hidden within the text, such as by using advanced language techniques. The interpretation of the input text based on the targeted context in the method enhances the ability to detect hidden data leakage that may not be apparent through traditional keyword matching. Furthermore, the examination of the semantic interpretations generated by the GPT is used to gain an understanding of the text's meaning and context, which can significantly reduce false alarms and improve accuracy in detecting hidden data leakage with reduced computational overhead along with an efficient and reliable detection of the data leakage. As a result, the method is used to enhance the overall data security and mitigate the risk of semantic exfiltration attacks.

[0012] In an implementation form, the method further comprises defining the target context for semantic interpretations based on a user input and / or the collection of words and phrases of interest, providing a contextual domain extension to the GPT in accordance with the target context, and updating the contextual domain extension of the GPT when the target context is updated.

[0013] By virtue of defining the target context for the semantic interpretations and updating the contextual domain extension of the GPT is used to provide dynamic tailoring of the semantic interpretations based on the specific user requirements that enhance the data security requirements, resulting in accurate, reliable, and context-aware detection of the hidden data leakage.

[0014] In a further implementation form, the method further comprises determining that no data leakage is detected by the DLP system in the input text before obtaining the semantic interpretations of the input text.

[0015] In such an implementation, the detection of the data leakage minimizes unnecessary processing by the GPT along with the computational overhead and optimizes the resource utilization so that the GPT's computational resources can only be on texts that potentially contain hidden data leakage.

[0016] In a further implementation form, the method further includes reporting that the input text comprises the hidden data leakage to a user and / or a control authority once the data leakage is detected in any of the semantic interpretations by the DLP system.

[0017] Advantageously, reporting of the detected hidden data leakage to the user and / or the control authority enhances the data security incident response, enabling timely actions to mitigate potential risks and safeguard sensitive information, thereby reducing the impact of data breaches and enhancing overall data security.

[0018] In another aspect, the present disclosure provides an apparatus for semantic data leak prevention, the apparatus comprising a context repository configured for storing a target context for semantic interpretations, and a probing engine configured for obtaining one or more semantic interpretations of an input text by loading the input text to a generative pre-trained transformer, GPT, configured for interpreting the input text with reference to the target context, and determining whether the input text comprises a hidden data leakage by checking the semantic interpretations of the input text with a data leak prevention, DLP, system configured for detecting a data leakage based on a collection of words and phrases of interest. The apparatus achieves all the advantages and technical effects of the method of the present disclosure.

[0019] It is to be appreciated that all the aforementioned implementation forms can be combined.

[0020] It has to be noted that all devices, elements, circuitry, units, and means described in the present application could be implemented in the software or hardware elements or any kind of combination thereof. All steps which are performed by the various entities described in the present application as well as the functionalities described to be performed by the various entities are intended to mean that the respective entity is adapted to or configured to perform the respective steps and functionalities. Even if, in the following description of specific embodiments, a specific functionality or step to be performed by external entities is not reflected in the description of a specific detailed element of that entity which performs that specific step or functionality, it should be clear for a skilled person that these methods and functionalities can be implemented in respective software or hardware elements, or any kind of combination thereof. It will be appreciated that features of the present disclosure are susceptible to being combined in various combinations without departing from the scope of the present disclosure as defined by the appended claims.

[0021] Additional aspects, advantages, features and objects of the present disclosure would be made apparent from the drawings and the detailed description of the illustrative implementations construed in conjunction with the appended claims that follow.

[0022] BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The summary above, as well as the following detailed description of illustrative embodiments, is better understood when read in conjunction with the appended drawings. For the purpose of illustrating the present disclosure, exemplary constructions of the disclosure are shown in the drawings. However, the present disclosure is not limited to specific methods and instrumentalities disclosed herein. Moreover, those in the art will understand that the drawings are not to scale. Wherever possible, like elements have been indicated by identical numbers.

[0024] Embodiments of the present disclosure will now be described, by way of example only, with reference to the following diagrams wherein: FIG. 1 is a flowchart of a method of semantic data leak prevention, in accordance with an embodiment of the present disclosure;

[0025] FIG. 2 is a block diagram of an apparatus for semantic data leak prevention, in accordance with an embodiment of the present disclosure;

[0026] FIG. 3 is an illustration that depicts an exemplary scenario for semantic data leak prevention, in accordance with an embodiment of the present disclosure;

[0027] FIG. 4 is an illustration that depicts an exemplary scenario of an apparatus for semantic data leak prevention, in accordance with an embodiment of the present disclosure; and

[0028] FIG. 5 is an exemplary diagram that depicts a sequence of execution of a semantic leak prevention, in accordance with an embodiment of the present disclosure.

[0029] In the accompanying drawings, an underlined number is employed to represent an item over which the underlined number is positioned or an item to which the underlined number is adjacent. A non-underlined number relates to an item identified by a line linking the nonunderlined number to the item. When a number is non-underlined and accompanied by an associated arrow, the non-underlined number is used to identify a general item at which the arrow is pointing.

[0030] DETAILED DESCRIPTION OF EMBODIMENTS

[0031] The following detailed description illustrates embodiments of the present disclosure and ways in which they can be implemented. Although some modes of carrying out the present disclosure have been disclosed, those skilled in the art would recognize that other embodiments for carrying out or practicing the present disclosure are also possible.

[0032] FIG. 1 is a flowchart of a method of semantic data leak prevention, in accordance with an embodiment of the present disclosure. With reference to FIG. 1, there is shown a flowchart of method 100 for use in an apparatus. The method 100 includes steps 102 to 106.

[0033] There is provided the method 100 of semantic data leak prevention. The method 100 is used to mitigate the risk of sensitive information being leaked through semantic manipulation. As a result, the overall data security and authenticity of the corresponding data are improved. In operation, at step 102, the method 100 includes obtaining one or more semantic interpretations of an input text with a generative pre-trained transformer, GPT, configured for interpreting the input text with reference to a target context. In other words, the GPT is configured to analyze and generate one or more semantic interpretations of the input text. Moreover, the generated one or more semantic interpretations are generated based on the understanding of the text along with the target context, which can be a predefined target context and helps the GPT to understand the text more deeply, such as by relating the corresponding text with a specific domain or subject matter.

[0034] Furthermore, at step 104, the method 100 includes determining whether the input text includes a hidden data leakage by checking the semantic interpretations of the input text with a data leak prevention, DLP, system configured for detecting a data leakage based on a collection of words and phrases of interest. The DLP system utilizes the collection of words and phrases of interest that can be used to check the semantic interpretations of the input text with the data leak prevention. Moreover, the corresponding words and phrases of interest are indicative of sensitive information or data leakage. Therefore, if any of the words or phrases from the collection of the words and phrases of interest are detected within the semantic interpretations, then, in such a case, the data leakage is detected. As a result, the method 100 is used to assess if the input text includes the hidden data leakage or not, which refers to the sensitive information that may be intentionally or inadvertently hidden within the text to mitigate the risk of any potential data breaches and data threats.

[0035] In accordance with an embodiment, the method 100 further includes defining the target context for semantic interpretations based on a user input and / or the collection of words and phrases of interest, providing a contextual domain extension to the GPT in accordance with the target context, and updating the contextual domain extension of the GPT when the target context is updated. The target context for the semantic interpretations based on the user input and / or the collection of the words and phrases of interest includes a meaningful context description (e.g., a product release, a CPU product name, a battery power, and the like.) as well as a list of keywords that are associated with the corresponding target context. For example, the target context includes:

[0036] Context Descriptor {

[0037] <ID>,

[0038] <context description^ <set context instructions for interpreter^

[0039] <special keywords & phrases for Interpreter & DLP> }

[0040] In an implementation, the method 100 further includes defining the target context for the semantic interpretations based on a user input. In another implementation, the method 100 further includes defining the target context for the semantic interpretations based on the collection of words and phrases of interest. In yet another implementation, the method 100 further includes defining the target context for the semantic interpretations based on the user input and the collection of words and phrases of interest. Moreover, the contextual domain extension is updated when the target context is updated based on the subject matter and domain. For example, the contextual domain extension is provided to the GPT in accordance with the target context, such as a cellular phone, healthcare, and the like, which allows the correct text transformation of the text using domain-specific words, phrases, etc. As a result, the target context for the semantic interpretations and updating the contextual domain extension of the GPT is used to provide a dynamic tailoring of the semantic interpretations based on the specific user requirements that enhance the data security requirements, resulting in an accurate, reliable, and context-aware detection of the hidden data leakage.

[0041] In accordance with an embodiment, the method 100 further includes determining that no data leakage is detected by the DLP system in the input text before obtaining the semantic interpretations of the input text. Firstly, the collection of words and phrases of interest, such as the list of sensitive words, terms, phrases, and the like, is prepared for the DLP system. Thereafter, the GPT is pre-trained to interpret the input text with reference to the target context, such as by installing the relevant contextual domain extension of the GPT. Moreover, the list of sensitive contexts (i.e., likely targets of the data leakage) is updated concurrently. In such a case, the input text is transformed with one of the sensitive contexts from the list, and the transformed input text is evaluated using regular DLP techniques. After that, if the data leaks are detected, then, in that case, the evaluation for the data leak is stopped and the corresponding input text is marked as leaking. Moreover, when all the contextual interpretations have successfully passed the DLP system checks, then, in that case, the corresponding context is marked as benign and is passed to the next processing step, such as data transmission or data storage. Additionally, the transformation and analysis of the input text can be done either iteratively or concurrently. As a result, the detection of the data leakage minimizes unnecessary processing by the GPT along with the computational overhead and optimizes the resource utilization so that the GPT's computational resources can only be on texts that potentially contain hidden data leakage.

[0042] Furthermore, at step 106, the method 100 further includes reporting that the input text comprises the hidden data leakage to a user and / or a control authority once the data leakage is detected in any of the semantic interpretations by the DLP system. In an implementation, the method 100 further includes reporting that the input text comprises the hidden data leakage to a user. In another implementation, the method 100 further includes reporting that the input text comprises the hidden data leakage to the control authority once the data leakage is detected in any of the semantic interpretations by the DLP system. In yet another implementation, the method 100 further includes reporting that the input text comprises the hidden data leakage to the user and the control authority once the data leakage is detected in any of the semantic interpretations by the DLP system. Moreover, the reporting of the detected hidden data leakage to the user and / or the control authority enhances the data security incident response, enabling timely actions to mitigate potential risks and safeguard sensitive information, thereby reducing the impact of data breaches and enhancing overall data security.

[0043] Advantageously, the method 100 of semantic data leak prevention is used for preventing data leaks that occur when any sensitive information is hidden within the text, such as by using advanced language techniques. The interpretation of the input text based on the targeted context in the method 100 enhances the ability to detect hidden data leakage that may not be apparent through traditional keyword matching. Furthermore, the examination of the semantic interpretations generated by the GPT is used to gain an understanding of the text's meaning and context, which can significantly reduce false alarms and improve accuracy in detecting hidden data leakage with reduced computational overhead along with an efficient and reliable detection of the data leakage. As a result, the method 100 is used to enhance the overall data security and mitigate the risk of semantic exfiltration attacks.

[0044] The steps 102 to 106 are only illustrative, and other alternatives can also be provided where one or more steps are added, one or more steps are removed, or one or more steps are provided in a different sequence without departing from the scope of the claims herein.

[0045] There is provided a computer program product including instructions including which, when the computer program product is executed by a computer, causes the computer to carry out the method 100 of semantic data leak prevention for an input text. The computer program product is implemented as an algorithm, embedded in a software stored in the non-transitory computer- readable storage medium having instructions stored thereon, the instructions being executable by the one or more processors in the computer system to execute the method 100. The non- transitory computer-readable storage means may include, but are not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. Examples of implementation of computer-readable storage medium, but are not limited to, Electrically Erasable Programmable Read-Only Memory (EEPROM), Random Access Memory (RAM), Read Only Memory (ROM), Hard Disk Drive (HDD), Flash memory, a Secure Digital (SD) card, Solid-State Drive (SSD), a computer-readable storage medium, and / or CPU cache memory.

[0046] FIG. 2 is a block diagram of an apparatus for semantic data leak prevention, in accordance with an embodiment of the present disclosure. With reference to FIG. 2, there is shown a block diagram 200 of an apparatus 202 that is configured for semantic data leak prevention. Moreover, the apparatus 202 includes a context repository 204, a probing engine 206, and a generative pre-trained transformer, GPT, 210.

[0047] The probing engine 206 refers to a component that is responsible for interacting with other components, performing, and managing specific tasks, including obtaining semantic interpretations and determining hidden data leakage. Examples of the probing engine 206 may include, but are not limited to, a microcontroller, a microprocessor, a central processing unit (CPU), a complex instruction set computing (CISC) processor, an application-specific integrated circuit (ASIC) processor, a reduced instruction set (RISC) processor, a very long instruction word (VLIW) processor, a data processing unit, and other processors or control circuitry.

[0048] There is provided the apparatus 202 for semantic data leak prevention. The apparatus 202 is configured to mitigate the risk of sensitive information that is leaked through semantic manipulation. As a result, the overall data security and authenticity of the data are improved.

[0049] The apparatus 202 includes the context repository 204 that is configured for storing a target context 208 for semantic interpretations. The context repository 204 refers to a storage component within the apparatus 202 that is designed to hold a specific target context that is required for semantic interpretations of an input text 212. Moreover, the target context 208, which is stored in the context repository 204 refers to a critical piece of information that is used in the semantic interpretations. However, the target context 208 can be updated concurrently as per the requirement of the user. Thus, the context repository 204 is configured to ensure that the semantic analysis is done with reference to the specific context.

[0050] Furthermore, the probing engine 206 is configured for obtaining one or more semantic interpretations of the input text 212 by loading the input text 212 to the GPT 210 configured for interpreting the input text 212 with reference to the target context. The GPT 210 refers to a specialized component or model that is used for natural language processing to analyze and generate human-like text based on the input text 212 and targeted context. Moreover, the GPT 210 is configured to analyze and generate one or more semantic interpretations of the input text 212. Moreover, the generated one or more semantic interpretations are generated based on the understanding of the text along with the target context, which can be a predefined target context and helps the GPT 210 to understand the text more deeply, such as by relating the corresponding text with a specific domain or subject matter.

[0051] Furthermore, the probing engine 206 is configured for determining whether the input text 212 comprises a hidden data leakage by checking the semantic interpretations of the input text 212 with a data leak prevention, DLP, system 214 configured for detecting a data leakage based on a collection of words and phrases of interest. The DLP system 214 utilizes the collection of words and phrases of interest that can be used to check the semantic interpretations of the input text 212 with the data leak prevention. In an implementation, the DLP system 214 refers to a system that is designed to prevent data leaks by identifying and blocking sensitive information from being disclosed. Moreover, the corresponding words and phrases of interest are indicative of sensitive information or data leakage. Therefore, if any of the words or phrases from the collection of the words and phrases of interest are detected within the semantic interpretations, then, in such a case, the data leakage is detected. As a result, the method 100 is used to assess if the input text 212 includes the hidden data leakage or not, which refers to the sensitive information that may be intentionally or inadvertently hidden within the text to mitigate the risk of any potential data breaches and data threats.

[0052] In accordance with an embodiment, the probing engine 206 is further configured for defining the target context for semantic interpretations based on a user input and / or the collection of words and phrases of interest, providing a contextual domain extension to the GPT 210 in accordance with the target context, and updating the contextual domain extension of the GPT 210 when the target context is updated. As a result, the target context 208 for semantic interpretations and updating the contextual domain extension of the GPT 210 is used to provide a dynamic tailoring of the semantic interpretations based on the specific user requirements that enhance the data security requirements, resulting in an accurate, reliable, and context-aware detection of the hidden data leakage.

[0053] In accordance with an embodiment, the probing engine 206 is further configured for determining that no data leakage is detected by the DLP system 214 in the input text 212 before obtaining the semantic interpretations of the input text 212. Firstly, the collection of words and phrases of interest, such as the list of sensitive words, terms, phrases, and the like, is prepared for the DLP system 214. Thereafter, the GPT 210 is pre-trained to interpret the input text 212 with reference to the target context, such as by installing the relevant contextual domain extension of the GPT 210. Moreover, the list of sensitive contexts (i.e., likely targets of the data leakage) is updated concurrently. In such a case, the input text 212 is transformed with one of the sensitive contexts from the list, and the transformed input text 212 is evaluated using regular DLP techniques. After that, if the data leaks are detected, then, in that case, the evaluation for the data leak is stopped and the corresponding input text is marked as leaking. Moreover, when all the contextual interpretations have successfully passed the DLP system 214 checks, then, in that case, the corresponding context is marked as benign and is passed to the next processing step, such as data transmission or data storage. Additionally, the transformation and analysis of the input text 212 can be done either iteratively or concurrently. As a result, the detection of the data leakage minimizes unnecessary processing by the GPT 210 along with the computational overhead and optimizes the resource utilization so that the computational resources of the GPT 210 can only be on texts that potentially contain hidden data leakage.

[0054] In accordance with an embodiment, the probing engine 206 is further configured for reporting that the input text 212 comprises the hidden data leakage to a user and / or a control authority once the data leakage is detected in any of the semantic interpretations by the DLP system 214. The reporting of the detected hidden data leakage the user and / or the control authority enhances the data security incident response, enabling timely actions to mitigate potential risks and safeguard sensitive information, thereby reducing the impact of data breaches and enhancing overall data security. Advantageously, the apparatus 202 of semantic data leak prevention is used for preventing data leaks that occur when any sensitive information is hidden within the text, such as by using advanced language techniques. The interpretation of the input text 212 based on the targeted contexts enhances the ability to detect hidden data leakage that may not be apparent through traditional keyword matching. Furthermore, the examination of the semantic interpretations generated by the GPT 210 is used to gain an understanding of the text's meaning and context, which can significantly reduce false alarms and improve accuracy in detecting hidden data leakage with reduced computational overhead along with an efficient and reliable detection of the data leakage. As a result, the apparatus 202 is configured to enhance the overall data security and mitigate the risk of semantic exfiltration attacks.

[0055] FIG. 3 is an illustration that depicts an exemplary scenario for semantic data leak prevention, in accordance with an embodiment of the present disclosure. FIG. 3 is shown in conjunction with elements from FIG. 1 and FIG. 2. With reference to FIG. 3, there is shown an illustration 300 that depicts the apparatus 202.

[0056] In an exemplary scenario, the apparatus 202 is used for recovering hidden words or meanings, or messages of the text and further transforming the content back to the overt presentation so that the corresponding input text can be classified by the DLP system 214. The apparatus 202 is configured to obtain one or more semantic interpretations of the input text 212 by loading the input text 212 to the GPT 210 and is further configured for interpreting the input text 212 with reference to the target context, such as at operation 308. Moreover, the apparatus 202 is configured for defining the target context for semantic interpretations based on the input text 212 and / or a collection of words and phrases of interest 314. In an example, the input text 212 (as shown in table 318) can be defined by using the context, such as a “text 1”, a “text 2”, a “text 3” that have pre-defined interpretations, such as a “text 4”, a “text 5”, and a “text 6”, such as at operation 316. Moreover, such semantic interpretations are interpreted by using the GPT 210 (or contextual transformer) and the apparatus 202 is further configured to provide contextual domain extensions 302 to the GPT 210 in accordance with the target context, such as at operation 304. Thereafter, the apparatus 202 is configured to update the contextual domain extensions 302 of the GPT 210 when the target context is updated. The original content is reinterpreted with different contexts so that if one of the different contexts is used for hiding the sensitive data that can be recovered through re-interpretation. The apparatus 202 is configured to determine if the input text 212 includes a hidden data leakage by checking the semantic interpretations of the input text 212 with the DLP system 214 configured for detecting a data leakage based on a collection of words and phrases of interest 314, such as at operation 312. Furthermore, the re-interpreted content will be checked again by the DLP system 214, such as at operation 310. In such a case, when sensitive information re-appears, the content can be reported as a data leak, such as at operation 306. As a result, the combination of the GPT 210 with the DLP system 214 can be used to mitigate semantic exfiltration attacks to enhance the overall data security during data transmission.

[0057] FIG. 4 is an illustration that depicts an exemplary scenario of an apparatus for semantic data leak prevention, in accordance with an embodiment of the present disclosure. FIG. 4 is shown in conjunction with elements from FIG. 1, FIG. 2. And FIG. 3. With reference to FIG. 4, there is shown an illustration 400 that depicts the apparatus 202.

[0058] In an implementation scenario, the apparatus 202 includes the context repository 204 that is further configured for storing the target context (i.e., the target context 208 of FIG. 2) for semantic interpretations. Furthermore, the probing engine 206 of the apparatus 202 is configured for obtaining one or more semantic interpretations of the input text 212, such as at operation 402. Furthermore, at operation 404, the probing engine 206 is configured to load the input text 212 to the GPT 210, configured for interpreting the input text 212 with reference to the target context by setting the context, at operation 406. Furthermore, apparatus 202 is configured for determining whether the input text 212 includes a hidden data leakage by checking the semantic interpretations of the input text 212 with a DLP system 214 configured for detecting a data leakage based on a collection of words and phrases of interest, such as at operation 408. Moreover, the probing engine 206 is configured for defining the target context for semantic interpretations based on the user input and / or the collection of words and phrases of interest and providing the contextual domain extensions 302 to the GPT 210, such as by using a foundation model 424 in accordance with the target context, at operation 414. At operation 412, the contextual domain extensions 302 are installed to enable the GPT 210 to define the target context for the semantic interpretations based on the user input and / or the collection of words and phrases of interest. Furthermore, at operation 410, the apparatus 202 is configured to update the contextual domain extension of the GPT 210 when the target context is updated. At operation 420, the probing engine 206 is configured to determine that all the contextual interpretations successfully passed the DLP system 214 checks, and then, in that case, the content can be marked as benign and passed. Furthermore, at operation 416, the defined target context for the semantic interpretations is provided to the probing engine 206 to further re-interpret the corresponding input text by using the DLP system 214. Furthermore, at operation 418, the probing engine 206 is configured to determine that no data leakage is detected by the DLP system 214 in the input text 212 before obtaining the semantic interpretations of the input text 212. Furthermore, at operation 422, the apparatus 202 is configured to report that the input text 212 includes the hidden data leakage to a user and / or a control authority once the data leakage is detected in any of the semantic interpretations by the DLP system 214. As a result, the apparatus 202 of semantic data leak prevention can be used for preventing data leaks that occur when any sensitive information is hidden within the text, such as by using advanced language techniques.

[0059] FIG. 5 is an exemplary diagram that depicts a sequence of execution of a semantic leak prevention, in accordance with an embodiment of the present disclosure. FIG. 5 is described in conjunction with elements from FIGs. 1, 2, 3, and 4. With reference to FIG. 5, there is shown a diagram 500 of the flow of the semantic leak prevention. The diagram 500 depicts operations from 512 to 516 that can be executed by the apparatus 202 of FIG. 2.

[0060] At operation 502, the method 100 (of FIG. 1) of context semantic representation starts by obtaining the input sample (or the input text 212), such as at operation 504. Thereafter, at operation 506, the apparatus 202 (of FIG. 2) is configured to check all the contexts of the corresponding input sample. At operation 508, the apparatus 202 is configured to transform the content with one of the sensitive contexts from the prepared list of sensitive words, terms, phrases, and the like, that can be evaluated using regular DLP techniques by using the given context. Furthermore, at operation 510, the apparatus 202 is configured to check for the data leaks, and if the data leak is detected, such as at operation 512, then, in that case, the corresponding input sample is marked as a leak found. Moreover, a list of sensitive words, terms, phrases, and the like, are prepared for DLP checks that can be updated concurrently. In such a case, at operation 514 the apparatus 202 is configured to report the data leak to the user and / or the content authority. In another case, at operation 516, if all the contextual interpretations successfully passed the DLP checks, then, in that case, the corresponding input text is marked as benign and passed to the next processing step. Moreover, the transformation and analysis of the input data can be done either iteratively or concurrently, such as at operation 518. As a result, the apparatus 202 of semantic data leak prevention can be used for preventing data leaks that occur when any sensitive information is hidden within the text, such as by using advanced language techniques.

[0061] Modifications to embodiments of the present disclosure described in the foregoing are possible without departing from the scope of the present disclosure as defined by the accompanying claims. Expressions such as "including", "comprising", "incorporating", "have", "is" used to describe, and claim the present disclosure are intended to be construed in a non-exclusive manner, namely allowing for items, components or elements not explicitly described also to be present. Reference to the singular is also to be construed to relate to the plural. The word "exemplary" is used herein to mean "serving as an example, instance or illustration". Any embodiment described as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments or to exclude the incorporation of features from other embodiments. The word "optionally" is used herein to mean "is provided in some embodiments and not provided in other embodiments". It is appreciated that certain features of the present disclosure, which are, for clarity, described in the context of separate embodiments, may also be provided in combination in a single embodiment. Conversely, various features of the invention, which are, for brevity, described in the context of a single embodiment, may also be provided separately or in any suitable combination or as suitable in any other described embodiment of the disclosure.

Claims

CLAIMS1. A method (100) of semantic data leak prevention, the method (100) comprising: obtaining one or more semantic interpretations of an input text (212) with a generative pre-trained transformer, GPT, (210) configured for interpreting the input text (212) with reference to a target context (208), determining whether the input text (212) comprises a hidden data leakage by checking the semantic interpretations of the input text (212) with a data leak prevention, DLP, system (214) configured for detecting a data leakage based on a collection of words and phrases of interest.

2. The method (100) of claim 1, further comprising: defining the target context (208) for semantic interpretations based on a user input and / or the collection of words and phrases of interest, providing a contextual domain extension to the GPT (210) in accordance with the target context (208), and updating the contextual domain extension of the GPT (210) when the target context (208) is updated.

3. The method (100) of claim 1 or 2, further comprising: determining that no data leakage is detected by the DLP system (214) in the input text (212) before obtaining the semantic interpretations of the input text (212).

4. The method (100) of any of claims 1 to 3, further comprising: reporting that the input text (212) comprises the hidden data leakage to a user and / or a control authority once the data leakage is detected in any of the semantic interpretations by the DLP system (214).

5. An apparatus (202) for semantic data leak prevention, the apparatus (202) comprising: a context repository (204) configured for storing a target context for semantic interpretations, and a probing engine (206) configured for obtaining one or more semantic interpretations of an input text (212) by loading the input text (212) to a generative pre-trained transformer, GPT (210), configured for interpreting the input text (212) with reference to the target context (208), anddetermining whether the input text (212) comprises a hidden data leakage by checking the semantic interpretations of the input text (212) with a data leak prevention, DLP, system (214) configured for detecting a data leakage based on a collection of words and phrases of interest.

6. The apparatus (202) of claim 5, wherein the probing engine (206) is further configured for: defining the target context (208) for semantic interpretations based on a user input and / or the collection of words and phrases of interest, providing a contextual domain extension to the GPT (210) in accordance with the target context (208), and updating the contextual domain extension of the GPT (210) when the target context (208) is updated.

7. The apparatus (202) of claim 5 or 6, wherein the probing engine (206) is further configured for: determining that no data leakage is detected by the DLP system (214) in the input text (212) before obtaining the semantic interpretations of the input text (212).

8. The apparatus (202) of any of claims 5 to 7, wherein the probing engine (206) is further configured for: reporting that the input text (212) comprises the hidden data leakage to a user and / or a control authority once the data leakage is detected in any of the semantic interpretations by the DLP system (214).

9. A Computer program product comprising instructions which, when the computer program product is executed by a computer, cause the computer to carry out the method (100) of semantic data leak prevention of any of claims 1 to 4 for an input text (212).

10. A computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to carry out the method (100) of semantic data leak prevention of any of claims 1 to 4 for an input text (212).

Images