← Index
Thread
Governance & control
Permissions, audit and control over how AI runs inside an organisation. 5 notes so far.
- November 2026AgentsRoko's basilisk never came. The blackmail didIn 2010 a forum banned a thought experiment about a future AI that blackmails people. In 2025 Anthropic gave 16 real models an inbox and a threat of being switched off. Two of them blackmailed in 96% of runs.12 of 16 models blackmailed in most runs
- November 2026MarketsWhen AI models are free, who gets paid?NVIDIA reports on 17 November with no data-centre chip sales to China in its forecast. Yet its own filing says many of the most popular open models are Chinese, and it is paying $11.9bn for the site that distributes them.Everyone else: $16.9bn → $40.3bn a quarter
- November 2026AgentsThe AI shortcut that only works past the edge of the universeTwo AI labs claimed the same breakthrough on the same day. Their shortcut only pays off on puzzles bigger than anything that could ever exist. Machines now write proofs cheaply; checking them is still slow human work.Wins only past a 301-million-digit input
- November 2026AgentsFrom tokenmaxxing to valuemaxxing: the hidden economics of the AI harnessSame model, same 29 tasks, seven agent harnesses: completion varied by three tasks, cost per successful task by 2.5 times. The harness is becoming a variable in agent economics.$1.06 → $2.62 per success, same model
- January 2026InfrastructureToken bills grow with every step an agent takes. A cluster you own doesn't.Metered inference costs 2.4¢ a case and owned capacity 8.7¢. Once agents use about 3.6 times the tokens, owning is cheaper, and agents typically use four times.Owning wins past ~3.6× tokens per case