As artificial intelligence (AI) systems become increasingly critical to national security and other high-stakes domains, the risk of model theft, manipulation, and misuse by sophisticated adversaries grows. In this report, the authors present a vertically integrated security strategy for secure inference data centers (SIDCs): purpose-built, Security Level 5 facilities designed to protect trained AI models against the most-advanced nation-state adversaries.

Growing concerns about the societal risks posed by advanced artificial intelligence (AI) systems have prompted debate over whether and how the U.S. government should promote stronger security practices among private-sector developers. Although some companies have made voluntary security commitments, competitive pressures and inconsistent approaches raise questions about the adequacy of self-regulation. At the same time, government intervention carries risks: Overly stringent security requirements could limit innovation, create barriers for small firms, and harm U.S. competitiveness. In this report, the authors help navigate these issues and identify a variety of practicable policy options for government and industry to strengthen frontier AI security.

If I could pick one piece of advice to shout from every rooftop to minimize the overall impact of ransomware, it would be: 📢GET YOUR BACKUPS OFF THE WINDOWS DOMAIN📢 Across MOXFIVE’s incident response work, almost all ransomware victims we work with have some sort of backup. This is great news – backups maybe the single strongest element to ransomware resilience. But during real cyber events, we find that backups are unusable 63% of the time. How alarming! These victims paid for a backup product – and developed a sense of confidence around that solution – only for the backups to fail when they are most needed.

We introduce entropy-based block selection (EBBS or “entropy triage”), a novel method to programmatically repair files rendered unusable after failed encryption processes. MOXFIVE has applied entropy triage while responding to ransomware attacks since 2023, repairing virtual machine disks that could not be decrypted with attacker-provided decryption tools because the prior malicious encryption process failed. The method measures Shannon entropy to select either pre-decryption or post-decryption data blocks to construct a usable file. Note that we simplify cryptographic descriptions for accessibility; using this method requires reverse engineers skilled with cryptography and malware.

Series of posts outlining the surprising costs of responding to a ransomware attack from investigation to litigation

“Attackers are already circling back to reselling stolen data instead of — and in addition to — extortion”

“Guest James Gimbi, Director at technical advisory firm MOXFIVE, shares why he believes banning ransomware payments is bad policy…”

As the policy community gets its arms around how government can address ransomware, we wanted to share an industry perspective on one proposal floated by thinkers outside the cybersecurity space: we believe ransomware payment bans are bad policy.

In today’s highly connected, interdependent world, all organizations rely on others for critical products and services. However, the reality of globalization, while providing many benefits, has resulted in a world where organizations no longer fully control—and often do not have full visibility into—the supply ecosystems of the products that they make or the services that they deliver. With more and more businesses becoming digital, producing digital products and services, and moving their workloads to the cloud, the impact of a cybersecurity event today is greater than ever before and could include personal data loss, significant financial losses, compromise of product integrity or safety, and even loss of life. Organizations can no longer protect themselves by simply securing their own infrastructures since their electronic perimeter is no longer meaningful; threat actors intentionally target the suppliers of more cyber-mature organizations to take advantage of the weakest link.

Brett Thorson and James Gimbi from BCG Platinion Cybersecurity share their thoughts on what the recent SolarWinds breach means for affected organizations, digital product and service providers, and their customers.

As awareness of cybersecurity supply chain risks grows among federal agencies, there is a greater need for tools that evaluate the impacts of a supply chain-related cyber event. This can be a difficult activity, especially for those organizations with complex operational environments and supply chains. A publicly available tool to support supply chain risk analysis that specifically takes into account the potential impact of an event does not currently exist. This publication describes how to use the Cyber Supply Chain Risk Management (C-SCRM) Interdependency Tool that has been developed to help federal agencies identify and assess the potential impact of cybersecurity events in their interconnected supply chains.

Covert communication is a rapidly expanding field of research with significant impact on the security theater. These communication methods, or “covert channels”, can be applied in a number of ways, including as a mechanism for an attacker to leak data from a monitored system or network. This paper sets out to contribute to this field by introducing a new covert channel which operates over transport layer protocols. The mechanism is flexible, covert, and has the potential to operate at relatively high bandwidth. In addition, this paper proposes a number of encoding schemes which can be used in conjunction with this channel to improve its bandwidth and covertness.