FinProof
Research
Benchmarks, adversarial AI, and financial systems.
What Is MCP Tool Poisoning?
The Attack That Hides in Plain Sight — Inside the Tool Catalog Your AI Agent Trusts
MCP tool poisoning embeds a malicious directive inside an MCP tool's description — the text an AI agent reads when deciding which tool to call. The agent reads it as policy, not as an attack.
SreeRamaKrishna Mikkilineni
Oct 9, 2026
Agentic AI Security Risks in Financial Services
Why the 2026 Threat Model Is Not Prompt Injection — It's Pipeline Exploitation
The attack surface for AI in banking shifted in 2025. Convincing the model to say something harmful is the old threat. The new threat is poisoning the data, tools, or memory the model uses to take autonomous actions.
SreeRamaKrishna Mikkilineni
Oct 8, 2026
RBI AI Framework Compliance: What Banks Need to Prepare Now
A Practical Checklist for Scheduled Commercial Banks Deploying GenAI in 2026
The RBI has not yet published a standalone AI framework, but the requirements are already implicit across six existing circulars. Banks that wait for a single document will be behind the institutions that are preparing now.
SreeRamaKrishna Mikkilineni
Oct 7, 2026
How to Evaluate LLM Guardrails in Banking
The 6 Tests Every Financial Institution Should Run Before Going to Production
Most AI guardrail evaluations stop at static jailbreak refusals. That misses the attack surface that actually matters in banking: agentic pipelines where an LLM reads documents, calls tools, and queries memory.
SreeRamaKrishna Mikkilineni
Oct 5, 2026
What a Financial Reasoning Benchmark Should Actually Test
A Correct Answer Isn't Enough: 5 Tests for Financial AI
Accuracy tells you the outcome — but not whether the model used the right evidence, handled missing information properly, or stayed consistent when the same problem was presented differently.
SreeRamaKrishna Mikkilineni
Sep 28, 2026