AI models don't kill people – people kill people
- ID
- 22950
- Status
- summarized
- Published
- 10 Sep 2026, 4:44 AM
- Fetched
- 10 Sep 2026, 5:22 AM
- Provider
- The Register
- Category
- technology
- Original URL
- https://www.theregister.com/ai-and-ml/2026/09/09/ai-models-dont-kill-people-people-kill-people/5295368
- Source URL
- https://www.theregister.com/headlines.atom
Summary
- Score
- 3.0
- Created
- 10 Sep 2026, 5:22 AM
- Tags
- Audience
- ai_ml_learnersai_agent_users
What happened
Anthropic researcher Jacob Coxon publicly resigned over concerns AI 'could kill us all by the end of the decade,' generating 110M+ views on X; his former colleague Evan Hubinger agreed, stating he personally estimates >10% chance of AI killing all humans within a decade and that Anthropic lacks a plan to solve superintelligence alignment. The Register's Thomas Claburn argues the real problem is irresponsible AI companies, not the models themselves, and dismisses existential risk framing by comparing it to climate change's 96,636 attributable heatwave deaths in 2023.
Why it matters
Opinion piece with no actionable technical or business takeaway for builders. The only concrete detail worth noting is that Anthropic's own science lead publicly states the company has no alignment plan for superintelligence and is 'not clearly on track' — relevant only if you are evaluating Anthropic as a vendor or thinking about AI safety policy positions.
Discussion angle
Whether internal Anthropic researchers publicly admitting no alignment plan exists should change how founders and developers think about deploying Claude-based agents in production systems.