What was checked
OpenAI in safety lockdown with internal model nearing AGI-level capability and narrow superintelligence; Anthropic Model 5 and Google Gemini 5 also approaching AGI-level with major leaps expected in next 3 months
Assessment summary
Needs cautionOpenAI reported pausing some reinforcement-learning training and strengthening monitoring and security controls, but it also said some workloads resumed and that research did not halt entirely. The available results reference future or numbered Anthropic models and benchmark discussion, but do not provide authoritative confirmation that a product called Anthropic Model 5 is approaching AGI.