“Information poisoning has by no means actually labored properly,” stated Bruce Schneier, chief of safety structure at Inrupt Inc., and a fellow and lecturer at Harvard’s Kennedy College. “Honeypots, no higher. This can be a intelligent thought, however I don’t see it as being something however an ancillary safety system.”
Joseph Steinberg, a US-based cybersecurity and AI advisor, disagreed, saying, “on the whole this might work for all types of AI and non-AI programs.”
“This isn’t a brand new idea,” he identified. “Some events have been doing this [injecting bad data for defense] with databases for a few years.” For instance, he famous, a database may be watermarked so whether it is stolen and a few of its contents are later used – a pretend bank card quantity, for instance — investigators is aware of the place that piece of information got here from. Not like watermarking, nonetheless, which places one unhealthy file right into a database, AURA poisons your complete database, so if it’s stolen, it’s ineffective.
