Black Hat and DEF CON are two of the major security conferences in the U.S., drawing large crowds of cyber and AI decision-makers to Las Vegas. Black Hat USA 2024 ran from Aug. 3-8, with most of the briefings occurring on Aug. 7 and 8; DEF CON 32 ran from Aug. 8-11. We’re rounding up the enterprise business tech news from Black Hat and DEF CON that is most relevant for IT and tech decision-makers.
CrowdStrike given ‘Epic Fail’ award
One of the traditions of DEF CON is the Pwnie Awards, an irreverent night where trophies are given out for both extraordinary successes and extraordinary failures. CrowdStrike’s global outage earned them the latter. The Pwnie Awards chose CrowdStrike early, about a week after the outage in July, and presented the trophy at DEF CON on Aug. 10. CrowdStrike President Michael Sentonas accepted the trophy in person.
How to hold generative AI accountable
A major topic of conversation and research at Black Hat was how to hold generative AI accountable in the case of hallucinations, misinformation, or follow-on effects from generated content.
At the one-day AI Summit (ticketed separately from the rest of Black Hat), experts discussed how to secure AI models and applications for enterprise use, as well as the use of AI in cyberattacks.
AI Village at DEF CON tasked a team of hackers with exploring how to detect and report AI flaws. This event is notable because both the vulnerabilities and the methods of reporting those vulnerabilities will be under scrutiny. Ideally, the lessons learned at this event will help AI vendors build frameworks for more thorough and accurate reporting.
DARPA and other government organizations had a significant presence at DEF CON, as they presented information on securing generative AI. The AI Cyber Challenge (AIxCC) Semifinal Competition tested hackers’ skills in securing critical infrastructure in a hypothetical, futuristic city.
Researchers from cloud security company Wiz put generative AI infrastructure to the test in their study of AI-as-a-service platforms. The team hacked Hugging Face and Replicate, leading generative AI hosting services, using “malicious models” to move laterally within the platform. That gave them a backdoor into private AI models, at which point they could gain information on proprietary weights, user prompts, and datasets. From there, they could launch supply chain attacks from the AI-as-a-service platform.
Patches and vulnerabilities identified
Many organizations at Black Hat and DEF CON announced patches and remarkable vulnerabilities at their briefings. See the complete list of DEF CON speakers for more.
Sonos speakers could be compromised, allowing attackers to listen in, two researchers from NCC Group revealed on Aug. 8. The exploit is made possible through the WPA2 Handshake encryption protocol, which can give an attacker remote access to the kernel. The researchers demonstrated turning a Sonos device into a “wiretap” and jailbreaking a Sonos Era-100 smart speaker.
Researchers Dennis Giese and Braelynn, a security consultant at Leviathan Security Group, detailed their work in discovering physical and side-channel attacks on Digilock and SAG smart lockers. This research is a reminder not to reuse secret PINs across critical devices like safes and phones.
Aqua Security announced on Aug. 7 that it had pinpointed a vulnerability in six AWS cloud services that could let attackers execute code remotely or take over accounts. Amazon has since shut that door. The problem was that S3 buckets for those six services — CloudFormation, Glue, EMR, SageMaker, ServiceCatalog, and CodeStar — had names with similar patterns. Because of this, attackers could guess names to plant malicious code in legitimate S3 buckets.
On Aug. 9, Amazon released the following unattributed statement: “AWS is aware of this research. We can confirm that we have fixed this issue, all services are operating as expected, and no customer action is required.”
Elsewhere at Black Hat, Zenity CTO Michael Bargury demonstrated how attackers can hijack Microsoft Copilot using indirect prompt injection and by poisoning RAG — a popular method for improving the accuracy of AI models.
In his briefing, Bargury highlighted the challenges generative AI presents to security teams, including remote code execution and “promptware.” He also recommended methods for locking down Copilot access against malicious actors, including people already inside the target company.
The security world is still working on standardized protection for AI
Cybersecurity service HackerOne identified a few trends in the intersection between generative AI and security:
- Generative AI helps threat actors attack at greater scales than before.
- Generative AI needs to be defined in ways that allow for greater standardization in security and governance.
- Open-source models are on-trend.
“The first step we need to take is creating and agreeing upon a set of common definitions,” Michiel Prins, cofounder of HackerOne, wrote in an email to TechRepublic. “We must ask: What is AI? Is it GenAI or LLMs? What about the ML solutions that have been around for decades? The space is riddled with unclear definitions, which makes it increasingly difficult for people to understand each other.”
Enhancing security intelligence
X-Ops, the security response team of IT-as-a-service provider Sophos, released a report on Tuesday about new tactics ransomware attackers use to put pressure on their victims. These tactics can include:
- Encouraging customers to open legal cases against victim organizations.
- Opening legal cases themselves.
- Seeking financial information about target companies, particularly information that might reveal inaccuracies or subterfuge.
- Exposing criminal activity that may occur on company devices.
- Painting the organizations they target as negligent or morally deficient.
Notable product releases
Flashpoint released new features and capabilities in Flashpoint Ignite and Echosec on Aug. 6. Flashpoint Ignite, the flagship platform, will now include investigations management and intelligence requirements mapping, which match Flashpoint collections with Priority Intelligence Requirements. Echosec will include location protection starting Aug. 6.
The AI security company CalypsoAI boosted its product line with out-of-the-box scanners for specific business-use cases and verticals and real-time threat updates.
Keynotes bring national and corporate players
Keynote speakers for Black Hat 2024 included Cybersecurity and Infrastructure Security Agency Director Jen Easterly, Google Security Engineering Manager Ellen Cram Kowalczyk, and Microsoft Threat Intelligence Strategy Director Sherrod DeGrippo.
TechRepublic covered Black Hat and DEF CON remotely.
