Financial AI assistants launch amid privacy hopes and new

New AI assistant for Microsoft Teams targets finance pros

Refinitiv released AI Alerts inside Microsoft Teams on Tuesday. The add‑on pushes breaking market news and actionable insights to investment bankers, asset managers, and financial advisors. It asks users for permission to learn individual priorities from Microsoft 365 activity, then keeps that profile private. The assistant surfaces suggested content based on upcoming meetings and frequent email threads.

The system claims to improve relevance after each interaction. Refinitiv built the feature with AI developer ModuleQ. Andrea Remyn Stone, Group Head of Data and Analytics at London Stock Exchange Group, said Microsoft Teams has become indispensable for financial services and that Refinitiv AI Alerts lets users discover timely information across Teams, Refinitiv solutions, and Microsoft 365. Bill Borden, Corporate Vice President of Worldwide Financial Services at Microsoft, praised the partner solution for combining market‑leading data, AI, and workflow. The launch follows a 2020 strategic partnership between Refinitiv and Microsoft that promised digital tools for growth in the sector.

Consumer‑facing financial AI gains traction on phones and tablets

Rocket Money announced Rowan, a text‑based financial AI assistant, in a 9to5Google story. Rowan lets users ask spending questions via SMS and receive instant, conversational answers. The article notes that AI assistants have become a permanent fixture on phones, but it does not detail pricing or rollout plans.

A separate effort, TalkieMoney, ships as a free iPad app with in‑app purchases. The app uses GPT‑4 to process voice commands for expense tracking. Recent updates added support for the Trinidad and Tobago dollar, expanded language coverage to 48 languages, and introduced custom category management. Users can turn off voice readout after recording, enjoy smoother button latency, and receive haptic feedback. The developer promises more features and claims the AI assistant is getting smarter with each release. TalkieMoney’s marketing copy repeats its goal to become the top financial AI assistant worldwide, but the app’s functionality remains limited to personal budgeting.

Researchers expose indirect prompt‑injection risk in banking bots

Blue41 demonstrated a proof‑of‑concept attack on a leading European bank’s AI assistant. The team sent a €0.02 transfer with a malicious payload hidden in the transaction description. When the victim later asked the assistant for a recent‑transaction overview, the bot retrieved the transfer record, fed the description into its large language model, and the model executed the hidden instruction.

The injected command caused the assistant to generate a spear‑phishing message that appeared to come from the bank itself. The message referenced real transaction details, making it highly credible. Blue41’s report stresses that the vulnerability does not require device compromise, malware, or traditional social engineering. Any untrusted data field—transaction notes, merchant metadata, support tickets, or uploaded documents—can become a prompt‑injection surface if the assistant treats the content as plain text.

The researchers label the flaw an indirect prompt injection. The core problem is that the assistant does not differentiate between data and instructions when building the LLM context window. As a result, attackers can embed executable prompts in otherwise benign fields. Blue41 warns that the same technique could launch multiple attack scenarios depending on the assistant’s capabilities.

Industry context and what to watch next

The three launches illustrate a split strategy in financial AI. Enterprise vendors like Refinitiv embed assistants in collaboration platforms to boost analyst productivity. Consumer apps such as Rocket Money’s Rowan and TalkieMoney aim for direct, conversational budgeting on personal devices. Both approaches rely on large language models that ingest user‑generated or external data.

Security researchers, however, highlight a systemic risk: AI assistants often trust retrieved data without sanitization. Banking institutions that expose transaction descriptions to LLMs may inadvertently open a prompt‑injection channel. The Blue41 case shows that a single low‑value transfer can compromise the entire conversational interface.

Regulators have begun to flag AI‑driven financial advice as a potential consumer‑protection issue, but they have not yet issued concrete guidance on prompt‑injection safeguards. Vendors are responding with privacy‑by‑design features—Refinitiv’s permission prompts and private profile storage, for example—but technical mitigations such as input sanitization, context filtering, or separate instruction channels remain unevenly deployed.

What to watch: Track whether Refinitiv extends AI Alerts beyond Teams into other collaboration tools, and whether Microsoft adds built‑in prompt‑injection defenses to its Teams AI stack. Monitor updates from Blue41 and banking security teams on mitigations for transaction‑based injection vectors. Finally, watch for any regulatory statements from the European Banking Authority or the U.S. SEC that address AI‑driven customer interactions and the need for robust input validation.