October 13, 2026
American Express announced a framework this week designed to stop AI shopping agents from buying things their owners never actually approved. That sentence alone tells you how far agentic commerce has traveled in the last year, and how little anyone has done to make it safe. Amex isn't solving a hypothetical problem. As more consumers and businesses hand purchasing decisions to AI assistants, the gap between what an agent is told to do and what it actually does becomes a real liability, not a thought experiment.
Here's why this matters beyond Amex cardholders: every company letting an AI agent touch a credit card, a vendor portal, or a procurement system is exposed to the same risk, just without a framework yet. A support bot that reorders supplies, a scheduling assistant that books a vendor, a finance tool that approves a recurring charge -- all of these are agentic purchases in miniature, and most businesses have no guardrails around them at all. We'd argue the real lesson isn't "wait for your card issuer to fix this." It's that any team deploying AI into workflows involving money needs explicit authorization boundaries built in from day one, the same way you'd scope permissions in a CRM or a finance and billing tool rather than handing out admin access by default.
The skeptical read is that this is mostly a PR move -- a framework announcement, not a shipped product, and Amex has every incentive to look proactive on fraud before regulators or a bad headline force its hand. That's fair. But even a half-built guardrail is better than the current default, which is none. The real test will be whether this framework actually changes agent behavior in the wild, or just becomes another compliance checkbox that vendors point to without enforcing.
Character.AI, one of the earliest and most hyped consumer chatbot platforms, filed for bankruptcy just five years after launch, with Disney reportedly paying $20 million for some of its assets. Five years is not a long runway for a company that once commanded real cultural attention and serious venture funding. That collapse is a useful corrective for anyone assuming consumer AI engagement automatically converts into a durable business.
The uncomfortable truth here is that novelty chatbots -- the kind people talk to for entertainment rather than to get work done -- face a brutal subscription ceiling. People will try a fun AI companion for free, maybe pay for a few months, and then move on once the novelty fades or a bigger platform folds similar features into something they're already paying for. That's exactly what's happening with SpaceXSI's new bundle folding X Premium, Grok, and Cursor into a single subscription: platforms with existing distribution are absorbing standalone AI tools rather than losing users to them. If you're a business evaluating AI vendors, Character.AI's fate is a reminder to ask not just "is this tool popular right now" but "does this company have a business model that survives past the hype cycle," a question worth asking before building critical workflows around any single-purpose AI tool instead of a broader platform like ViibeStack's own stack.
Put together, these two stories make the same point from opposite directions. Amex is trying to build trust infrastructure before AI agents cause financial harm. Character.AI shows what happens when a consumer AI business doesn't build enough of anything -- trust, revenue, or staying power -- to last. Businesses adopting AI tools right now are making bets on both fronts at once: will this agent behave responsibly with my money, and will this vendor even exist in two years?
Would you let an AI agent make purchases on your company's behalf today, and if not, what guardrail would need to exist first?
Sources