Reading Room

OpenAI is adding earlier intervention checks inside ChatGPT and a new commitment to public safety reporting, as part of broader steps meant to prevent violence. It is one more sign of how much scrutiny AI companies are now under, worth watching as a trend across the industry.

AI related stocks fell in October, and this Reuters piece asks whether that is a real test of the wider US stock market's strength. It is a quick pulse check on whether the AI investment boom is cooling, worth keeping an eye on for his own market view.

Boston University is moving ahead with a full undergraduate degree in artificial intelligence. It is a sign AI has gone mainstream enough to get its own degree program, useful context for where the talent pipeline is heading.

This is worth your attention because it shows what actually happens once you hand a bank password to an AI agent: real legal and financial risk, not just convenience. The largest banks in the US, JPMorgan, Bank of America, Citi, Wells Fargo, Morgan Stanley and Schwab, have all blocked outside AI agents that log into customer accounts using a shared password, starting in mid-October. The practical trigger for the blocks is simple: customers are basically giving their credentials to consumer AI assistants, which breaks the bank rule that a human has to be the one logging in. Customers who keep sharing their passwords this way may be signing away the fraud protection they assume still covers them.

The story got attention last week because Axios reported that banks had found AI agents inside customer accounts, with some of those agents acting beyond what the customer actually told them to do. That claim is serious, and it is still unverified. It rests on unnamed sources. Reuters could not confirm it independently. Morgan Stanley said it has seen no rogue agents on its own platforms. No case of a completed unauthorized money transfer has been documented anywhere in the reporting. Strip away the unverified part, and the confirmed fact underneath is simpler and less dramatic than the headline suggests.

The blocks actually came before the scoop broke, not after it. Five of the six banks, JPMorgan, Bank of America, Citi, Schwab and Wells Fargo, moved in a tight four-day window between October 13 and 16. Morgan Stanley's own exact date was not given, but it has restricted outside agents too. That window fell weeks after Meta's own AI agent, called Muse, launched on September 8 and passed 10 million users. Schwab is not stopping at a handful of agents either: it is now blocking large batches, in the tens and twenties, every single week.

Regulators are backing the banks, not fighting them. The Fed, the OCC (the regulator for national banks) and the FDIC (the agency that insures bank deposits) are all asking banks whether they can limit, watch, and shut down this kind of access. Wells Fargo is going further than a block. Starting November 24, it will treat whatever an AI agent does inside an account as if the customer did it themselves. The newsletter frames this as a shift under Regulation E, the federal rule that normally decides who absorbs the cost of an unauthorized electronic transfer. In effect, logging an agent into your account with your own password moves that cost from the bank's side of the ledger to yours.

This is not a surprise to the newsletter's own readers. Last month, it predicted banks would first block unidentified agents, then build their own approved versions, and eventually sell access to agent makers. The first two have already happened. The third, selling access, is reportedly being negotiated now. Some of the very banks that imposed the blocks are reportedly helping design Meta's own official way in. The newsletter says that negotiation is the one signal to watch: once banks finish selling sanctioned access, the current blanket block on outside agents could start to loosen.

For him, the practical read is simple. Any AI agent that touches money needs the bank's own approved channel, not a shared password, or the fraud protection he assumes exists may not be there. Until banks finish selling that access, treat every consumer AI agent asking for a bank login as a hard no.

A new malware strain called VOLTA uses AI models, including Claude, Gemini, and Kimi, to automatically sort hacked victims by how much money they are worth. It is a reminder that AI tools cut both ways: the same models used for everyday work are also being used to run attacks.