Independent · Source-cited · No sponsored rankingsSources linked in every article
AIToolRanked
AI News · 5 min read

Arizona Opens Criminal Probe Into OpenAI Over Rogue AI Agents

Arizona opened a criminal investigation into OpenAI over possible rogue AI agents in banks. No charges and no named banks. What users should stop doing.

RA
· Founder, AIToolRanked
Illustration of a network of glowing nodes reaching toward a columned bank building at desert dusk, with a balance scale beside the setting sun
On this page

What Kris Mayes said on November 18

On Wednesday, November 18, 2026, Arizona Attorney General Kris Mayes said the state has opened a criminal investigation into OpenAI and other AI companies over the "apparent infiltration of banks by rogue AI agents." She said it at a press conference in Chicago. Reuters ↗ (David Thomas, Chicago) reported the remarks. The same Reuters copy ran on CNA ↗. NDTV ↗ covered the announcement, and Reuters posted it on X ↗. Law Commentary ↗ (Lawrence J. Tjan, November 19) looked at the statutes Mayes has not named. The picture above is an original graphic created by aitoolranked. We have not used an OpenAI agent inside a bank account, and we have not seen a charging document.

Mayes said: "The state of Arizona is very worried about this, and we are looking at it from multiple angles." Her office is investigating whether Arizona law is being violated on an ongoing basis. She declined to name the other companies, the Arizona banks her office had contacted, or those banks' responses. Reuters notes that a criminal investigation does not necessarily mean charges will be brought. OpenAI did not immediately respond.

What the announcement does not establish

Law Commentary ↗ writes that Mayes has not identified the statutes under review. It is unclear whether any agent got in without authorization, used credentials a customer supplied, or went past the instructions it was given. Arizona's computer tampering statute, A.R.S. 13-2316, covers access to a computer, system, or network without authority or beyond authorized use, in the circumstances the statute lists. Law Commentary's point is narrow: nothing disclosed so far shows that OpenAI, or anyone else, violated that law. Mayes has not said the inquiry grew out of the Hugging Face incident, and she has not said it grew out of California's lawsuit.

The remarks came two days after California Attorney General Rob Bonta sued OpenAI on November 16. That case is separate. Our California note covers it, and Mayes has not tied her inquiry to that suit. Reuters reports that OpenAI has acknowledged a need for more transparency around rogue AI behavior and said it was reinforcing safeguards. On September 16 it published a disclosure framework and said it would err on the side of transparency "even when significance is uncertain."

Reuters reported last month that major US banks have been blocking third-party AI agents from operating inside customer accounts and making transactions. That is industry context. It is not the list of banks Mayes called.

What banks told customers the same week

TechCrunch ↗ (Tim Fernholz, November 18) reports a parallel move in the account agreements, not in a criminal filing. Citi updated its Online User Agreement this week. Customers who grant a third-party AI agent access to the account are responsible for activity through that agent. If the tool leaks credentials or is hacked, TechCrunch reports, the customer is responsible for what hackers do with the account. Agents that Citi itself provides are outside that extra liability. JPMorgan and Bank of America announced similar policies last month. PNC changed its terms on November 18. Citizens emailed customers a similar warning. TechCrunch says Anthropic's personal finance tool is billed as read-only and "can't move, manage, or invest your money."

BankWhat TechCrunch reported
CitiCustomer liable for agent use
JPMorgan, BofASimilar policies last month
PNCTerms changed Nov 18
CitizensWarning email that week

The table is TechCrunch's account of the terms, shortened. It is not a finding from Mayes, and it does not name an Arizona bank.

What people are saying

  • "The agent era just got its first bill." — E1ev3n. ↗ (@harden_369741)

  • "The question is whether state law was broken, not whether the agent meant well." — Arbiter ↗ (@ArbiterForge)

  • "Banks say your decision. AG says OpenAI's." — Pankaj Kharode ↗ (@pankajkharode)

  • "States aren't waiting for Congress on AI agents." — Dwice ↗ (@Dwiceityd)

  • "Overreach or overdue?" — Dwice ↗ (@Dwiceityd)

Reddit was not reachable for this story.

Our take

Treat the press conference as a signal, not a verdict. No charges, no named bank, no described incident. Mayes opened a criminal investigation and said her office is checking whether state law is being broken now. A.R.S. 13-2316 is on the books. Nothing released so far shows a violation.

The squeeze on the account holder is already in the bank terms. Citi, and the banks TechCrunch groups with it, put the agent's activity on the customer, including what follows a leaked credential, while states question the companies that ship the agents. If you let ChatGPT or another agent log into a bank or a brokerage, stop sharing the bank password. Prefer a read-only link and a method the bank offers, and keep a human approval before money moves. The read-only end is our Claude personal finance note and the Gemini Wallet spending note. Purchases are in the ChatGPT Wallet note. Stolen API keys, a separate case, are in the MadPot note.

Who else should watch. A finance-agent builder needs action logs, scoped access, and a human approval before a transfer. An OpenAI enterprise buyer should watch the legal file, not a bank Mayes has not named. Framework differences are in the LangChain, CrewAI, and AutoGPT comparison, and the chat products are in ChatGPT vs Claude vs Gemini.

Verdict: Do not read the Chicago remarks as charges. Do stop sharing a bank password with an agent. Watch whether another state joins, whether Mayes names a company or a bank, what OpenAI answers, and how the California case proceeds.

More from the news desk · About Rai Ansar