Editorial rule: documented fact, inference, and speculation remain separate. A source appearing here does not mean its claims are adopted as fact beyond what the source itself establishes.
Current primary-source watchlist
Dartmouth - AI at 70: What Must Remain Human?
Official 2026 conference framing around human judgment, responsibility, and accountability in an AI-saturated world.
METR - Task-Completion Time Horizons of Frontier AI Models
Current measurements of how long and complex a task frontier AI agents can complete at defined reliability levels. The page was updated in 2026 and supersedes earlier 2025 snapshots.
Anthropic - Alignment Research
2026 work includes scalable oversight, automated alignment research, disempowerment patterns, and updated investigations of alignment failures.
Research papers under refresh
The historical 1956-2025 papers remain valuable for background, but documents ending in 2025 are now treated as historical snapshots. The 2025-2059 forecasting paper and frontier-safety synthesis require the highest-priority 2026 update because capability, oversight, alignment, governance, and agent research have continued to move rapidly.
No refreshed 2026 paper will be presented as authoritative project research until it has been reviewed and accepted under the 2059 project governance process.