AIニュース Oct 2, 2026

今日の注目点

• AI governance and ethicsThe White House's 'super intelligence' executive order and AI safety pledge, alongside Anthropic's reported consciousness discussions with religious leaders, show governments and companies are now framing AI in existential terms, not just productivity terms.
• AI deployment and securityOpenAI's GPT-6 deployment guide and Dots agent platform launch, combined with Apple's new full disk access limits, indicate the industry is shifting from model capability to production safety and workflow integration as the key competitive battleground.
• AI in cybersecurityGreg Kroah-Hartman's finding that only 20 of 79 AI-generated vulnerability reports required fixes highlights how AI-assisted security scanning may create noise that distracts maintainers, even as Kevin Mandia's Armadin raises $255.5 million for agent-based security testing.
• AI and professional workA Mercor study showing AI models outperform licensed CPAs on structured accounting tasks, while no model completes all APEX Benchmark tasks unsupervised, suggests AI is reshaping professional work unevenly—strong in defined workflows, weaker in complex judgment scenarios.
• AI company internal dynamicsOpenAI firing three researchers over alleged leaks to an outside safety organization, per the Wall Street Journal, comes as the company publishes deployment guides and launches agents, underscoring tension between rapid commercialization and internal safety transparency.

AI 生成分析