BASI: the outsiders testing the inside
BASI adds a different kind of power to this map: the practical ability to make AI systems do things their developers did not intend. That knowledge can travel through a public community, a specialist collective, a research engagement or a commercial tool. Following those routes gives a clearer picture than assigning the entire neighborhood to a single ideological camp.
Pliny’s project hub connects public tools, BT6 and the BASI community. A Latent Space interview presents Pliny and John V’s case for open research. John V’s dated roster identifies operators with other professional roles. These are distinct circles of activity, not interchangeable names for a proven corporate hierarchy.
The institutional links are tangible. OpenAI credits Lysios for external testing. TrojAI lists John V with Lysios. a16z names Pliny as a grant recipient. Grants, professional roles and testing credits are the working connections here.
What does the system achieve? Public experimentation produces methods and skilled practitioners. Organized research gives that work a setting. Institutional engagements can give researchers resources and access, while developers gain testing capacity and a public account of external scrutiny. This is our interpretation of the documented relationships; the actual agreements would tell us how publication rights and competing interests are negotiated.
One technical claim deserves care. In Basilisk Venom, Pliny and Marco Figueroa interpret elicited model behavior as evidence of training-data ingestion. It is an attributed research claim, not a recovered training log or proof of a network intrusion.
What to follow next
The most revealing missing records are grant terms, research agreements and public explanations of disclosure rights. For the collective itself, dated membership statements and named project credits would clarify who performed which work. Those records would show how much autonomy survives when independent researchers enter institutional relationships.