November 19, 2026AgentsAgent-Operable

Banks to Agent Users: If It Gets Hacked, You Pay

Citibank changed its online user agreement this week with a blunt message for anyone letting a third-party AI agent into their account: whatever the agent does is your responsibility, and if the tool leaks your credentials or gets hacked, so is whatever the hackers do. PNC changed its terms on Wednesday, Nov 18, and Citizens emailed customers with the same warning, TechCrunch reported. JPMorgan and Bank of America made similar moves last month. https://techcrunch.com/2026/11/18/more-banks-says-customers-using-ai-agents-are-on-their-own-even-if-they-get-hacked/

Look at the fine print and the strategy is clear. Citi's terms note that AI agents Citi itself provides can be used without that extra liability. So the bank is not saying agents are dangerous. It is saying other people's agents are your risk and its own agents are its product. Anthropic's new personal-finance tool, for comparison, is explicitly read-only and cannot move money.

The same day, OpenAI's site carried an essay called The Silent Consensus, adapted from a Harvard lecture and labeled as the author's own views, not OpenAI's. No byline is shown on the page. Its target is the unstated plan of handing companies and institutions to AI one reasonable delegation at a time, with no rebellion required. Its fix is plumbing: people must be tied to the actions of their agents, and we can build agents whose permissions are intelligible and limited, whose actions can be audited, and whose authority users can withdraw. https://openai.com/index/the-silent-consensus At the DealBook Summit, Sam Altman and Greg Brockman made the same case on stage, arguing that people need to stay in control and that a gradual handover to AI must not happen. https://techcrunch.com/2026/11/18/sam-altman-warns-against-a-future-where-ai-not-humans-calls-the-shots/

Put those side by side and you see the gap. The essay describes an authority layer that does not exist yet: scoped, auditable, revocable permissions. The banks, lacking it, did the only thing a legal department can do overnight and pushed all the risk to the user.

That is the opening for whoever builds the missing layer. The first agent that can prove to a bank exactly what it was allowed to do, and that it did nothing else, gets the liability clause rewritten in its favor.
← Previous
Inside the Codex Sandbox: OpenAI Shows the Lock and the Gap
Next β†’
A 19-Day Ban Tested Coding Benchmarks. Merge Rates Did Not Move
← Back to all articles

Comments

Loading...
>_