Hackers are always improving their ability to break through network defense in order to steal valuable documents. Therefore, some researchers suggest using artificial intelligence algorithms to hide the files that need to be protected in a large number of realistic fake files to completely confuse the intruders. At present, the online false repository generation engine (we-forge) based on word embedding can generate false patent documents under research“ In the future, it may be able to generate false copies of any files you want to protect. ” V. s. subrahmanian, the developer of the engine and a network security researcher at Dartmouth College, said.
If hackers want to steal the formula of a new drug, they will have to look for fake prescriptions like looking for a needle in a haystack. This means that hackers need to examine each equation in detail, and it may lead to a dead end“ In this way, the difficulty of the game increases, “Subramanian said.” it will make those who steal feel painful. ”
Subramanian said that when he traced some cases, he found that companies on average knew all this 312 days after they were attacked by the new network“ Hackers have nearly a year to steal documents and intellectual property, “he said.” even Pfizer, hackers have enough time to steal it. Not only the Pearl on the crown, but also other ‘jewelry’ and ‘watches’ will be stolen. ” Rachel Tobac, CEO of social proof security, a network security company (not involved in the study), said that forged documents generated by we-forge can also be used as hidden “trigger lines”. For example, a decoy file can issue a security warning when accessed. The company’s usual strategy is to make fake documents manually“ But now, if AI can do this, we just need to create a lot of fake files to confuse hackers. ” Tobak said.
The system will generate convincing bait files by searching the keywords of the files. For each keyword found, it calculates a list of related concepts and replaces the original term with a randomly selected word. This process produces dozens of documents that do not contain proprietary information and still look very credible. Subramanian and team members also invited graduate students majoring in computer science and chemistry to evaluate the true and false patents in their respective fields. In their opinion, the files generated by we-forge are very credible. The relevant research was published in the journal Transactions on management information systems of the American computer association.
We-forge may expand its application scope in the future, but Subramanian also pointed out that the documents of action guide are much more difficult than those of technical formula. Nevertheless, both he and tobak believe that the research has commercial value“ I absolutely believe that there will be companies that can apply this type of products, “tobak said.” if artificial intelligence technology creates credible bait files without disclosing sensitive details, I think it will be successful. “