Bravo
If ARM was an arm, BRN would be its biceps💪!
Oooooh...I wonder if BrainChip has a solution to this problem?
CYBERSECURITY
By: Israel Ojoko
Published: January 16, 2024 at 4:46 pm EST
A significant vulnerability, named ‘LeftoverLocals,’ has been uncovered in graphics processing units (GPUs), impacting major manufacturers including Apple, Qualcomm, and AMD. This revelation comes from researchers at the security firm Trail of Bits, who discovered that this vulnerability could potentially allow attackers to extract up to 180 megabytes of data from a GPU’s memory. This could include sensitive data such as queries and responses generated by large language models (LLMs) and the weights that drive the response.
A Flaw in Millions of Apple, AMD, and Qualcomm GPUs Could Expose AI Data
Patching every device affected by the LeftoverLocals vulnerability—which includes some iPhones, iPads, and Macs—may prove difficult.
www.wired.com
CYBERSECURITY
LeftoverLocals: A Significant Vulnerability in Major GPUs
By: Israel Ojoko
Published: January 16, 2024 at 4:46 pm EST
A significant vulnerability, named ‘LeftoverLocals,’ has been uncovered in graphics processing units (GPUs), impacting major manufacturers including Apple, Qualcomm, and AMD. This revelation comes from researchers at the security firm Trail of Bits, who discovered that this vulnerability could potentially allow attackers to extract up to 180 megabytes of data from a GPU’s memory. This could include sensitive data such as queries and responses generated by large language models (LLMs) and the weights that drive the response.
GPUs: A Security Blind Spot
GPUs were not originally designed with the same degree of data privacy as central processing units (CPUs). This has created a security vulnerability that allows for the potential exfiltration of sensitive information. To exploit ‘LeftoverLocals,’ an attacker would need some level of operating system access on the target’s device. The vulnerability disrupts the data silos which prevent users sharing the same processing resources from accessing each other’s data. The researchers demonstrated a proof of concept attack where a response from an LLM was extracted from vulnerable GPU memory within seconds using less than 10 lines of code.Coordinated Disclosure and Mitigation Efforts
After testing 11 chips across seven GPU makers, the vulnerability was disclosed in September through a coordinated effort with the US CERT Coordination Center and the Khronos Group. Apple, Qualcomm, and AMD have confirmed the impact, with Apple releasing fixes in its latest M3 and A17 processors, Qualcomm providing firmware patches, and AMD planning to offer fixes in March as optional mitigations. Google has also released fixes for ChromeOS devices with impacted AMD and Qualcomm GPUs.Challenges and Implications
However, distributing these patches to end users is challenging due to the multiplicity of parties involved in the tech ecosystem. The implications of this vulnerability are significant, as chaining multiple vulnerabilities together is common in sophisticated digital attacks. Gaining initial access to a device is often a prerequisite for various types of cyberattacks, thus making the ‘LeftoverLocals’ vulnerability a potential vector for these malicious activities.
Last edited: