A safety floor that each lab can move on its own isn't a floor
29 October 2026 (Tech Policy Press). The French non-profit CeSIA commented on AI companies repeatedly rewriting their own safety thresholds, in a 2026 Tech Policy Press article on measuring frontier AI capabilities.
Canada and Europe Need a Way to See the AI They Cannot Access
Who else?
What it refers to: AI companies set their own safety capability thresholds and have quietly revised them, as Anthropic did with its biological weapons threshold.
Other voices
- “The companies are writing the rules” Jen Rosiere Reynolds,
- “They're gonna police themselves, it's going to work out very well” Donald Trump, Politician,
- “This is no substitute for hard law” David Robinson, Researcher,
- “Why do the mistakes systematically tend to guard and favour Anthropic's interests?” Nathan Calvin,
- “Airlines around the world actually do not compete on safety” Andrew Ng, Builder,










