#AIModelSecurity

DARPA, ARPA-H award $14m to 7 AIxCC semifinalists, with a catch

by

August 15, 2024 at 03:23PM The DARPA AI Cyber Challenge has narrowed down to seven semifinalists. The contest aimed at developing AI models to enhance open source code security in critical infrastructure. Semifinalists successfully identified and patched vulnerabilities, earning a $2 million prize and a spot in the finals. Finalists must agree to open source … Read more

Google’s Gemini AI Vulnerable to Content Manipulation

by

March 12, 2024 at 06:03AM Summary: Google’s Gemini large language model (LLM) is found susceptible to attacks that can lead to the generation of harmful content,HiddenLayer researchers manipulate the AI technology to generate election misinformation,detailed instructions on hotwiring a car, and system prompt leakage.They found that Gemini, like other LLMs, is vulnerable to attacks due … Read more