It's actually not that bad.

People on welfare can already borrow money on credit cards and so on. If they get into default the only legally enforceable repayment arrangements are ones where they are not forced below subsistence levels. Yes, lenders can end up getting pennies a week. Yes, it is basically their fault,

You can get the situation where someone borrows against their livelihood in some kind of libertopia where the lenders right to their money overrides the borrowers right to eat.

Also, if someone is using GBI to start a business, borrowing to buy equipment is pretty reasonable,

The GBI is a guaranteed income stream, right? So, can I sell it? Can I put it up as a collateral for a loan?

It's up to the willingness of the country giving out the loan to regulate whether it wants to enforce those loans. Many countries have a legal system where you can't collect money from people who need that money to live.

GiveDirectly and the direct transfer RCTs in Africa/Third World countries don't answer the question about First World poverty because almost everyone, including the industrious and drug-free and high functioning people, in those countries is dirt-poor; in the First World, there is a much stronger correlation of pathology and poverty. To give an example, the direct transfers in Africa work because people there really are in poverty traps where $100 can make a big difference in letting them buy a cow or a motorcycle, and this is why the direct transfer RCTs show benefits; no one in America will show big benefits from a few transfers of $100 because poor people there have problems which can't be solved by some cash.

The upcoming YC-funded experiment will help test the generalizability of basic income results, and the original American experiments decades ago suggest that a basic income wouldn't cause lots of self-destructive behavior (or at least, wouldn't make things noticeably worse), but on the other hand, the natural experiments of lotteries in the USA and elsewhere like Sweden show minimal benefits to random shocks of wealth (which could've been invested for income). So I wouldn't be totally pessimistic, but I also wouldn't be surprised if BI experiments in the USA do worse than one would predict from the earlier GiveDirectly results.

I've had with poor people don't bear this out.

Are you talking about poor people, or the fellow dying in the street. There's a difference.