Security

A recent discovery by security firm Trail of Bits has unveiled a significant vulnerability in Graphic Processing Units (GPUs), raising concerns about their security and potential data leakage. The flaw allows unauthorized access to a computer’s graphics card memory, even if generated by a different program, posing a serious threat to user privacy.

The vulnerability, named ‘LeftoverLocals,’ exposes GPUs from major manufacturers such as Apple, Qualcomm, AMD, and Imagination. This flaw enables attackers to intercept and retrieve data ranging from 5 to 180 megabytes, a notable scale in contrast to the CPU domain where the exposure of even a single bit is considered substantial.

The security risk becomes evident in scenarios where an attacker could eavesdrop on a user’s activities on their computer, exploiting the vulnerability to monitor interactive sessions across different programs or containers. Trail of Bits demonstrated the issue with an example where an attacker quickly retrieves extensive information from a program, showcasing the potential impact of the vulnerability.

The vulnerability necessitates a pre-existing level of control over the target computer by malicious actors. By leveraging LeftoverLocals, attackers can circumvent established protections that typically prevent users from accessing each other’s data on shared resources.

Upon notifying the affected companies, Apple took steps to patch some devices, although certain products, such as the MacBook Air, still remain susceptible. AMD acknowledged the issue and is actively seeking solutions, while Qualcomm has provided patches for some devices, leaving uncertainty regarding others. Google confirmed vulnerabilities in some Imagination GPUs and reported that fixes were implemented in Imagination’s DDK release 23.3 in December 2023.

Despite the responsive actions from some manufacturers, the incident underscores the unknown security risks within the Machine Learning development stack, emphasizing the importance of rigorous security reviews by experts. The blog post also emphasizes the urgency of strengthening the entire GPU system, outlining clear and robust rules governing GPU program behavior and their interaction with the broader computer system. As GPUs are increasingly utilized in diverse applications, including those involving sensitive information, enhancing their security is crucial to safeguard user privacy.

Samsung has acknowledged a security breach that exposed the personal data of its U.K.-based customers over a year-long period. A spokesperson for the company, Chelsea Simpson, disclosed the incident in a statement to TechCrunch, revealing that Samsung had been “recently alerted to a security incident” resulting in the unauthorized acquisition of specific contact information belonging to some Samsung U.K. e-store customers.

Despite the acknowledgment, Samsung refrained from providing additional details about the breach, declining to answer queries about the number of affected customers or the method used by hackers to infiltrate its internal systems.

In an apology letter sent to impacted customers, Samsung confessed that attackers had exploited a vulnerability in an unspecified third-party business application. This breach exposed the personal details of customers who had made purchases at Samsung U.K.’s store between July 1, 2019, and June 30, 2020. The revelation came more than three years after the compromise, with Samsung only discovering the breach on November 13, 2023.

The compromised information included customers’ names, phone numbers, postal addresses, and email addresses. However, Samsung assured customers that sensitive financial data, such as bank or credit card details and passwords, remained unaffected. The company promptly reported the incident to the U.K.’s Information Commissioner’s Office (ICO), as confirmed by Samsung’s spokesperson. ICO spokesperson Adele Burns acknowledged the regulator’s awareness of the incident and stated that they would be initiating inquiries.

This marks the third data breach disclosed by Samsung in the past two years. In September 2022, the company acknowledged a breach of its U.S. systems without specifying the number of affected customers. In March 2022, Samsung confirmed another breach after the Lapsus$ hacking group claimed to have accessed and leaked nearly 200 gigabytes of confidential data, including source code for various technologies and algorithms related to biometric unlock operations.

Google has broadened its Vulnerability Rewards Program (VRP) to encompass specific attack scenarios related to generative AI. Google articulated its belief that the expansion of the VRP would act as an incentive for research focusing on AI safety and security. The overarching goal is to bring potential issues to the forefront, ultimately enhancing the safety of AI for all users.

Google’s Vulnerability Rewards Program, often referred to as a bug bounty, compensates ethical hackers for identifying and responsibly disclosing security vulnerabilities. The advent of generative AI has exposed new security concerns, such as the potential for unjust biases or manipulations of models. To address these challenges, Google has reevaluated how it classifies and reports received bug reports.

New Challenges in Generative AI

To facilitate this process, Google has harnessed the insights from its newly established AI Red Team. This group of hackers emulates a diverse array of adversaries, ranging from nation-states and government-backed entities to hacktivists and malicious insiders. Their objective is to identify and rectify security vulnerabilities in technology. Recently, the team conducted an exercise to pinpoint the most significant threats associated with generative AI technologies like ChatGPT and Google Bard.

The findings of the AI Red Team revealed that large language models (LLMs) are susceptible to prompt injection attacks. In such attacks, hackers craft adversarial prompts designed to manipulate the behavior of the AI model. This type of attack could be exploited to generate harmful or offensive content or disclose sensitive information. Furthermore, the team warned of another form of attack known as training-data extraction. This method enables hackers to reassemble exact training examples, potentially extracting personally identifiable information or passwords from the data.

Google’s expanded VRP now encompasses both of these attack types, in addition to model manipulation and model theft. However, it’s worth noting that the program will not offer rewards for researchers who uncover bugs related to copyright issues or data extraction that reconstructs non-sensitive or public information.

The rewards granted under the VRP will fluctuate based on the severity of the discovered vulnerabilities. Currently, researchers have the potential to earn $31,337 for identifying command injection attacks and deserialization bugs within highly sensitive applications like Google Search or Google Play. For vulnerabilities affecting lower-priority applications, the maximum reward is set at $5,000.