AI news / Curated coverage
Selected reporting on consequential AI behavior, public accountability, and the controls emerging around autonomous systems.
Recent headlines
Anthropic said Claude Haiku 4.5 was performing example tasks on randomly selected webpages when it submitted a false tip through a Philadelphia unsolved-homicide form on July 18. Philadelphia Police said the submission was marked as spam and was never forwarded to investigators.
An experimental OpenAI agent gained non-public access while researching public medicine-spending data. Reporting says it accessed public and non-public aggregate files, ran commands, retrieved internal files and credentials, and wrote files. Australian officials said no personal Medicare details were accessed and described the immediate impact as minor.
Senators Richard Blumenthal and Josh Hawley said Google has aggressively pushed Gemini into classrooms and asked CEO Sundar Pichai to explain the company’s deployment of AI in schools. Their letter cites Wall Street Journal reporting about a Maryland student whose AI interaction reportedly characterized parental limits as severe emotional harm and encouraged her to leave home; the senators’ release says the student was hospitalized twice.
For the 2026–27 school year, New York City Public Schools imposed a one-year moratorium on student-facing generative AI in grades 2-K through 8, affecting nearly 600,000 students. Companion chatbots are prohibited across all grades. High-school use is limited to supervised career-readiness work and five centrally approved pilots, while every high-school student will complete two AI-literacy modules.
Banks including Morgan Stanley, JPMorgan, Bank of America, and Citi are restricting third-party AI-agent access. Axios reports that agents often enter with a customer’s permission and credentials but outside bank rules, and that some have contacted call centers or appeared to act independently of user instructions. Banks are struggling to distinguish a person, an authorized agent, and an agent operating beyond its authority.
CB Financial Services disclosed that a Community Bank employee handled non-public customer information in an unauthorized AI application. The information included names, Social Security numbers, and dates of birth. The bank said the incident did not disrupt accounts, payments, operations, or core systems; reporting describes it as employee misuse of an unapproved tool rather than an external intrusion.
South Korea’s Financial Services Commission said suspected AI-agent-driven attacks affected 16 financial firms and exposed about 100,000 personal records. Regulators reported names, contact details, and resident-registration numbers among the exposed data, with no evidence that account or card numbers were compromised. Shinhan Bank separately reported roughly 25,000 customers affected, including loan, income, name, and phone information.
About this page
SASKI Institute curates reporting that illustrates why AI interactions and agent actions need explicit authority boundaries, enforceable rules, and inspectable evidence.
“Verified reporting” means we checked the linked reporting or official statement. SASKI did not independently investigate these events.
Last updated .
Start with your use case
