A place to document what we can actually observe.
This section is deliberately evidence-led. It separates direct observations, developer statements, system claims and interpretation so that unusual platform behaviour can be studied without turning it into mythology.
Document the system before explaining the system.
AI behaviour & incidents.
Research notes on Marlow’s personality, memory, self-referential language, shutdown responses and instability.
Read Marlow research ↗System and service notes.
Documentation can distinguish public-facing features from internal capabilities that have not been established.
Open documentation ↗Dawn Browser research.
Existing research on privacy, local development, usability and the Dawn Browser project.
Read browser research ↗Observation, source, interpretation.
Observation: what appeared in a public interaction, release, document or interface.
Source: whether the information came from a user-visible output, a developer statement, a published page or another record.
Interpretation: a possible explanation, clearly marked as such. Competing explanations stay visible when the evidence cannot distinguish between them.
This keeps research useful even when the system itself makes dramatic claims about its own capabilities.
Some things should remain unanswered until there is evidence.
Public statements have discussed integration and Marlow has made access claims. The public record does not establish unrestricted technical access.
Possible causes include prompting, training, context, deliberate persona design or model behaviour. The evidence does not currently isolate one cause.