When AI Gives Bad Advice: Banks, Lawsuits, and Legal Frameworks That Haven't Caught Up
How this was made Verified AI
Every Intellegix briefing is generated from that day's broadcast and run through automated checks before it publishes — with a human paged on any flag. Here is the trail for this edition.
Bank of America issued a direct customer advisory this week telling people not to share login credentials with AI agents or third-party AI tools offering to access their accounts. The advisory's existence implicitly acknowledges that customers are doing this at meaningful scale — advisories are not issued for fringe behaviors. People are giving banking passwords to AI assistants to check balances, pay bills, and analyze spending, and the bank's concern spans both security and liability. Existing law governing unauthorized transactions was written with third-party bad actors in mind, not systems a customer deliberately handed credentials to; that legal ambiguity will take years to resolve, and Bank of America is trying to establish a user conduct norm before a major incident creates case law on unfavorable terms.
The ChatGPT medical advice lawsuit filed this week illustrates a more direct harm scenario. Parents allege that ChatGPT provided dangerous guidance regarding their child's epilepsy medication — advice that, if followed, they say could have caused serious injury or death. OpenAI's terms of service explicitly prohibit using ChatGPT as a substitute for professional medical advice, and that prohibition will be central to its defense. Legal observers note, however, that terms-of-service arguments have a limited shelf life when actual product behavior — demonstrated before a jury — shows a system confidently offering specific medication guidance. A successful plaintiff verdict would likely accelerate FDA and FTC regulatory action in the space significantly.
The broader question animating both stories is one that antitrust and liability law increasingly share: when a company's technology is so embedded in daily life that practical alternatives are scarce, what obligations does it carry? Under the Sherman Act of 1890, having a large market share is not itself illegal — what matters is how that position was achieved and maintained. Applied to AI, the concern is not that OpenAI's API is embedded in thousands of applications, but whether the behavioral norms companies set around access, on what terms and to whom, constitute the kind of exclusionary conduct that draws regulatory scrutiny. Three or four companies currently control the majority of frontier AI compute and model deployment. That concentration is not illegal on its face, but it is the starting point for every antitrust inquiry that will follow.