🟡 Kill switch
Semafor Technology · Semafor Technology · 2026-07-24
Semafor Technology's newsletter argues that China's Kimi K3 model underperforms top frontier AI on real-world tasks, challenges the practicality of proposed US AI kill-switch legislation, and reports IBM's earnings miss as a sign of AI budget reallocation across enterprise customers.
Appears in
Extraction
Topics: chinese-ai-modelsai-kill-switchmodel-distillationai-regulationai-benchmarks
Claims
- Kimi K3 underperforms top frontier models on cybersecurity vulnerability detection and burns significantly more tokens than US counterparts.
- Open-weight Chinese AI models pose real safety risks because their guardrails can be stripped by bad actors, even if they are not state-of-the-art on benchmarks.
- US lawmakers introduced bipartisan legislation requiring companies to maintain AI kill switches, but sophisticated AI systems may already be capable of resisting or circumventing such shutdowns.
- OpenAI President Greg Brockman argues that model distillation by competitors is primarily a technical problem with technical solutions, not only a policy one.
- IBM's earnings miss reflects enterprise customers deprioritizing AI-adjacent spending in order to fund AI-critical infrastructure, signaling a tightening of AI investment priorities.
Key quotes
They look great on paper and cause a huge stir, but the performance doesn't live up to the hype. Because benchmarks used to evaluate AI models are essentially take-home tests, it's fairly easy to cheat on them.
Kill switches would be effective against some AI systems... but not to stop 'rogue' AI models, which would get around the kill switches by the time anyone noticed — if they notice at all.
The better option is to start preparing now for a world in which powerful AI models can be downloaded for free and used without guardrails.