OpenAI admits new models likely to pose 'high' cybersecurity risk (www.techradar.com)

OpenAI has issued a warning regarding the potential cybersecurity risks associated with its future large language models (LLMs), which could be capable of developing zero-day exploits and aiding cyber-espionage efforts. Acknowledging the rapid advancements in AI capabilities, OpenAI views these developments as a double-edged sword, presenting both challenges and opportunities for cybersecurity professionals. To address this concern, the company is investing in strengthening its models for defensive tasks and enhancing capabilities for users engaged in cybersecurity, including improved auditing and vulnerability patching tools. To further safeguard against misuse, OpenAI plans to establish the Frontier Risk Council, comprising experienced cybersecurity experts who will advise on the balance between beneficial AI capabilities and potential threats. This council is expected to play a critical role in shaping the company’s cybersecurity strategy as it shares insights and best practices within the industry. OpenAI aims to proactively mitigate vulnerabilities by establishing robust access controls and monitoring systems while collaborating with other organizations to address the broader implications of AI capabilities in cybersecurity.