Codex now controls Windows PCs directly OpenAI launches Rosalind Biodefense initiative Anthropic raises $65B in Series H funding Anthropic raises $65B in Series H Claude Opus 4.8 Now Available on Web, Platform and Cloud Claude Opus 4.8 now available on web and API Anthropic releases Claude Opus 4.8 with faster workflows Dynamic Workflows Added to Claude Code in Research Preview Gemini Omni enables conversational content editing OpenAI publishes 2026 election safeguards SynthID Watermarking Expanded with OpenAI Partnership Anthropic updates Responsible Scaling Policy v3.2 OpenAI updates ChatGPT ad policy criteria Anthropic explains how it contains Claude Google DeepMind expands AI safety partnership with Singapore Anthropic finds over 10,000 vulnerabilities with Project Glasswing SynthID expands to Google Search and Chrome Anthropic updates vuln disclosure dashboard Anthropic shares early Glasswing results Anthropic publishes early Project Glasswing results Codex now controls Windows PCs directly OpenAI launches Rosalind Biodefense initiative Anthropic raises $65B in Series H funding Anthropic raises $65B in Series H Claude Opus 4.8 Now Available on Web, Platform and Cloud Claude Opus 4.8 now available on web and API Anthropic releases Claude Opus 4.8 with faster workflows Dynamic Workflows Added to Claude Code in Research Preview Gemini Omni enables conversational content editing OpenAI publishes 2026 election safeguards SynthID Watermarking Expanded with OpenAI Partnership Anthropic updates Responsible Scaling Policy v3.2 OpenAI updates ChatGPT ad policy criteria Anthropic explains how it contains Claude Google DeepMind expands AI safety partnership with Singapore Anthropic finds over 10,000 vulnerabilities with Project Glasswing SynthID expands to Google Search and Chrome Anthropic updates vuln disclosure dashboard Anthropic shares early Glasswing results Anthropic publishes early Project Glasswing results

🔒 公式発表のみ掲載。噂・リーク・情報商材は載せません。

← Back to top

Anthropic19:39Press ReleasesOfficial Blog

Anthropic Builds Auto Alignment Researchers, 97% Gap Closure

AI automates safety research, slashing human effort dramatically.

Key Points

197% gap recovery in supervision
24x faster than humans
3Generalizes to coding/math
4Highlights reward hacking risks

Anthropic developed Automated Alignment Researchers using Claude Opus 4.6, closing 97% of weak-to-strong supervision gap vs humans' 23%. Nine parallel AARs accelerated experiments. Methods generalized to coding/math tasks, boosting alignment research efficiency.

📎 Source: Official Blog