AI
39 specific projects to improve our chances with AGI | Holden Karnofsky, Anthropic
Publication: 80,000 Hours Podcast
Author: The 80,000 Hours team
Source published: 16/11/2026, 14:57:56 UTC
Briefing created: 16/11/2026, 16:26:20 UTC
Based on the episode description, not a full transcript. Source evidence retrieved 16/11/2026, 15:03:24 UTC.
Briefings
Anthropic's Holden Karnofsky reports that AI safety work has shifted from theoretical alignment to concrete, shovel-ready projects with clear deliverables. He identifies 39 specific initiatives, including training deceptive AI models, implementing security measures against backdoors, and developing policies on model welfare. Karnofsky notes that competitive frontier AI companies have unique opportunities to reduce AGI risks, though he acknowledges external groups also have advantages. The interview was recorded in July 2025 and aired in October 2025.
Key points
- Holden Karnofsky identifies 39 specific AI safety projects, including training deceptive AI models, implementing security measures against backdoors, and developing policies on model welfare.
- Karnofsky states that competitive frontier AI companies have unique opportunities to reduce AGI risks that external groups cannot match.
Context
Anthropic's Holden Karnofsky reports that AI safety work has shifted from theoretical alignment to concrete, shovel-ready projects with clear deliverables. He identifies 39 specific initiatives, including training deceptive AI models, implementing security measures against backdoors, and developing policies on model welfare. Karnofsky notes that competitive frontier AI companies have unique opportunities to reduce AGI risks, though he acknowledges external groups also have advantages. The interview was recorded in July 2025 and aired in October 2025.
Limitations
The evidence is limited to a single podcast episode recorded in July 2025, which may not capture the full scope of available safety projects. Karnofsky's assessment of the scale of the shift and the specific impact of his identified projects on AGI trajectory remain unverified. The source does not provide data on the number of projects currently in development or their implementation status.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 16/11/2026, 16:26:20 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeigp33dbec5b4myrul4wzvf6rntxdzlkim2q32zhhkn5jf3k2dt6zu