Join top executives in San Francisco on July 11-12 to hear how leaders are integrating and optimizing AI investments for success. Find out more
One of the most effective ways to test the security of an application is through the use of opponent’s attacks. In this approach, security researchers actively attack technology — in a controlled environment — to try and find previously unknown vulnerabilities.
It’s an approach currently supported by the Biden-Harris administration to help protect artificial intelligence (AI). Be a part of it Action to promote responsible AI announced yesterday, the administration has called for public reviews of existing general AI systems. As a result, this year’s DEF CON 31 security conference, to be held on August 10–13, will offer a general AI public review at AI village.
“This independent activity will provide important information to researchers and the public about the impact of these patterns, and will enable AI companies and developers to take steps to remediate incidents detected in those models,” the White House said in a statement. liberate, release, free.
DEF CON villages have a history of advancing security knowledge
The DEF CON security conference is one of the largest gatherings of security researchers in any given year and has long been the site of discovery and disclosure of new security vulnerabilities.
This won’t be the first time a village at DEF CON has targeted a technology that’s making nationwide headlines. In previous years, especially after the 2016 US election and concerns about election interference, Vote Villages were established at DEF CON in an effort to consider confidentiality (or lack of confidentiality). confidentiality) in voting machine technologies, infrastructure and processes.
With villages at DEF CON, attendees can discuss and explore technologies under a responsible disclosure model to help improve overall security. With AI, there is a particular need to examine the risks of the technology as it is more widely deployed in society at large.
How synthetic AI hacks will work
Sven Cattell, founder of AI Village, commented in a declare that, traditionally, companies have solved risk identification using specialized red teams.
The red team is a type of cybersecurity team that simulates attacks to detect potential problems. According to Cattell, the challenge with generalized AI is that a lot of the work around general AI has taken place in private without the red team’s judgment.
“The diversity problems with these models will not be solved until more people know how to form red teams and evaluate them,” says Cattell.
In terms of specifics, AI Village’s general AI attack simulation will include on-site access to large language models (LLMs) from participating vendors. The event will feature a chess points system approach, where attackers gain points for achieving certain goals that will demonstrate a range of potentially harmful activities. The individual with the highest score wins the “Premium Nvidia GPU”.
The review platform on which the event will take place is being developed by AI scale. “As the use of platform models becomes widespread, it is important to ensure that they are carefully evaluated for reliability and accuracy,” said Alexandr Wang, founder and CEO of Scale, told VentureBeat.
Wang noted that Scale spent more than seven years building the AI system from scratch. He claims that his company is also unbiased and does not depend on any single ecosystem. As a result, Wang says Scale can independently test and evaluate systems to ensure they are ready to be deployed into production.
“By bringing our expertise to a wider audience at DEF CON, we hope to ensure that advancements in the capabilities of the platform model happen alongside progress in the future,” said Wang. evaluation and safety of the model”.
VentureBeat’s Mission is a digital city square for technical decision-makers to gain knowledge of transformative and transactional enterprise technology. Explore our Briefings.