Just a reminder on existential safety ratings with the Pentagon news.

The US Department of Defense is proceeding with integrating advanced AI models, specifically xAI's Grok, into its classified networks to analyze sensitive military data. This strategic shift comes after AI safety-focused company Anthropic reportedly declined to modify its core constitutional AI safeguards to accommodate Pentagon requirements, sticking to its principles on AI development ethics. The military's turn to Grok, created by Elon Musk's xAI, marks a significant escalation in the operational deployment of frontier AI within national security infrastructure, directly contravening warnings from leading AI researchers like Geoffrey Hinton about the dangers of militarizing artificial intelligence.

The controversy is amplified by Grok's poor performance on key safety benchmarks. According to the Future of Life Institute's comprehensive AI Safety Index from Summer 2024, Grok scored only 1.5 out of 5 in the critical 'Existential Safety' category, which evaluates risks like power-seeking behavior and catastrophic misuse potential. This low rating, based on an assessment across six safety categories, places Grok far behind models like Anthropic's Claude and OpenAI's GPT-4 in formal safety evaluations. The Pentagon's deployment signals a prioritization of capability over caution, setting a precedent for how governments might adopt AI systems that commercial developers deem too risky for certain applications, potentially accelerating an AI arms race.