Automated authentication based on shared data
View Patent ↗Disclosed are various approaches for using generative large language models (LLMs) and a shared text corpus to facilitate automated in-depth authentication of a user. A private text corpus can be shared between the user account and a verification entity. At least one authentication prompt can be generated using a large language model trained at least in part on the private text corpus. At least one authentication prompt can be sent to an authentication agent associated with the user. The authentication agent can send an authentication response based at least in part on the at least one authentication prompt and the private text corpus. The authentication response can be verified based at least in part on the private text corpus.
1 . A system, comprising:
a computing device comprising a processor and a memory; and
machine-readable instructions stored in the memory that, when executed by the processor, cause the computing device to at least:
obtain a private text corpus associated with a user account, the private text corpus being shared between the user account and a verification agent associated with the computing device;
generate at least one authentication prompt using a large language model, the large language model being trained at least in part on the private text corpus;
send the at least one authentication prompt to an authentication agent associated with the user account, wherein the authentication agent is fine-tuned based at least in part on the private text corpus and is trained to identify a parsed term in the at least one authentication prompt;
receive, from the authentication agent, an authentication response, the authentication response being based at least in part on the parsed term, the at least one authentication prompt, and the private text corpus; and
verify the authentication response based at least in part on the private text corpus.
2 . The system of claim 1 , wherein the machine-readable instructions which cause the computing device to obtain the private text corpus further cause the computing device to at least:
receive a request for access from a client device, the request for access comprising at least one user credential;
identify the user account associated with the client device based at least in part on the at least one user credential; and
obtain, from a data store, the private text corpus associated with the user account.
3 . The system of claim 2 , wherein the machine-readable instructions further cause the computing device to at least grant access to the client device upon successful verification.
4 . The system of claim 1 , wherein the machine-readable instructions further cause the computing device to at least send, to a client device associated with the user account, a notification that verification was successful.
5 . The system of claim 1 , wherein the machine-readable instructions further cause the computing device to at least:
send an initiation request to the authentication agent, the initiation request identifying the verification agent;
receive an initiation response from the authentication agent; and
in response to receiving the initiation response, send the at least one authentication prompt to the authentication agent.
6 . The system of claim 1 , wherein the machine-readable instructions which cause the computing device to verify the authentication response further cause the computing device to at least:
generate an expected response to the at least one authentication prompt using the large language model; and
compare the authentication response to the expected response.
7 . The system of claim 1 , wherein the private text corpus is shared between a client device associated with the user account and the verification agent.
8 . A method, comprising:
sharing, by a client device, a private text corpus with a verification service hosted by a verification agent;
training, by the client device, a large language model based at least in part on the private text corpus;
incorporating, by the client device, the large language model into an authentication agent associated with the client device; and
training the authentication agent to identify a parsed term in an authentication prompt and to generate an authentication response to an authentication prompt using at least a parsed term and the large language model.
9 . The method of claim 8 , further comprising:
receiving, by the client device, a request to configure automatic authentication;
in response to receiving the request to configure automatic authentication, sending, by the client device, a request for the large language model; and
receiving, by the client device, the large language model.
10 . The method of claim 8 , wherein training the large language model further comprises:
converting, by the client device, the private text corpus into a plurality of vector embeddings;
storing, by the client device, the plurality of vector embeddings in a vector database; and
augmenting, by the client device, the large language model with the vector database.
11 . The method of claim 8 , further comprising:
encrypting, by the client device, the large language model; and
saving, by the client device, the large language model to a local storage location on the client device.
12 . The method of claim 11 , further comprising:
receiving, by the client device, an authentication initiation request;
sending, by the client device, a decryption notification to a user interface of the client device;
receiving, by the client device, a decryption input; and
in response to receiving the decryption input, decrypting, by the client device, the authentication agent.
13 . The method of claim 8 , wherein training the large language model further comprises fine-tuning, by the client device, the large language model based at least in part on the private text corpus.
14 . The method of claim 8 , wherein the private text corpus is shared between a user account and the verification agent.
15 . A non-transitory, computer-readable medium, comprising machine-readable instructions that, when executed by a processor of a computing device, cause the computing device to at least:
obtain a private text corpus associated with a verification agent, the private text corpus being shared between the verification agent and a user account associated with the computing device;
receive at least one authentication prompt from a verification service hosted by the verification agent;
identify a parsed term from the authentication prompt;
generate an authentication response corresponding to the at least one authentication prompt using an authentication agent, wherein the authentication agent generates the authentication response based at least in part on the parsed term, the authentication prompt, and a large language model, the large language model being trained at least in part on the private text corpus; and
send the authentication response to the verification service.
16 . The non-transitory, computer-readable medium of claim 15 , wherein the machine-readable instructions which cause the computing device to obtain the private text corpus further cause the computing device to at least:
receive a request for credentials from the verification service, the request for credentials identifying the verification agent; and
identify the private text corpus associated with the verification agent.
17 . The non-transitory, computer-readable medium of claim 15 , wherein the machine-readable instructions, when executed, further cause the computing device to at least receive a notification that verification was successful.
18 . The non-transitory, computer-readable medium of claim 15 , wherein the machine-readable instructions, when executed by the processor, further cause the computing device to at least:
receive an initiation request from the verification agent;
decrypt the large language model; and
upon decryption of the large language model, send an initiation response, the initiation response indicating that the large language model is ready for authentication.
19 . The non-transitory, computer-readable medium of claim 18 , wherein the machine-readable instructions which cause the computing device to decrypt the large language model further cause the computing device to at least:
send a decryption notification to a user interface; and
receive a decryption input via the user interface.
20 . The non-transitory, computer-readable medium of claim 18 , wherein the machine-readable instructions which cause the computing device to decrypt the large language model further cause the computing device to at least obtain a decryption input from a data store.