|
|
|
Photo by WenPhoto/Adobe Stock, design by Haley Okuley/RAND
|
|
Protecting advanced AI models from theft and misuse is becoming increasingly important. One key aspect of safeguarding these models is securing AI model weights—the numeric variables that encode the core intelligence of an AI, representing everything a model has learned.
A new RAND report looks at how AI model weights can be protected against a variety of potential malicious attackers, from opportunistic criminals to highly resourced nation-state operations. The study also identifies dozens of methods of attack, ranging from more mundane threats, such as basic social engineering schemes, to more severe measures, such as a military takeover.
The authors highlight actions that AI organizations can take right away to help prepare for such threats, including reducing the number of people authorized to access AI model weights, implementing insider threat programs, and engaging in third-party red-team exercises to simulate relevant threat actors.
These insights provide a first-of-its-kind playbook to help AI companies defend against the most-sophisticated kinds of attacks. The findings may also help facilitate meaningful engagement between policymakers, AI developers, and other stakeholders as they work to understand the broader impacts of AI security.
|
|
|
More on This Research
|
In this Q&A, the study’s lead author, Sella Nevo, discusses how AI models work, why securing AI model weights and systems is important, and more details on the research team’s key findings.
|
|
|
|
|
|
Fentanyl and other potent, illegally manufactured synthetic opioids have ravaged North America. As the use of synthetics in Europe increases, how can leaders there prevent a similar crisis from unfolding? In a new RAND Europe paper, researchers emphasize that proactive actions are crucial. These might include wastewater surveillance to monitor consumption of illegal opioids, updating overdose-response protocols to adjust for synthetic opioids’ greater potency, and strengthening prevention programs. “By acting in advance and thinking about the problem as a 'poisoning crisis,'” the authors conclude, “Europe has a priceless opportunity to prepare and implement a sensible response.” Read more »
|
|
|
|
Russian Defense Minister Sergei Shoigu was the latest senior military official to be relieved of his position by Vladimir Putin. Why would Putin oust a longtime loyalist with deep roots in the Russian military establishment? “Apparently,” says RAND's Hunter Stoll, “no amount of loyalty or experience can assuage Putin if the desired battlefield outcomes are not being delivered.” But regardless of the reasons for Shoigu's removal, this shake-up will likely lead to changes in Moscow's approach to its war in Ukraine. Read more »
|
|
|
|
|
|
