← Map · Transaction & Payment Processing Services · S&P 500
XYZ BLOCK CLASS A $45.6B
No. Fraud detection is live. The number is potential scam dollars, not a loss rate.
8 of 15
Reach, Weight, and Proof.
What they are doing
Block's second-quarter 2026 shareholder letter, filed with the SEC, says machine learning has been in production from the start because instant seller onboarding, rapid fraud detection, and loan underwriting do not work without it. The new AI chapter in that letter is internal: agents that write code, plus Moneybot and Managerbot. On May 4, 2026 the company said real-time machine-learning models score transactions in flight, and that Cash App payment warnings have marked more than $3.5 billion in potential peer-to-peer scams since August 2020.
Why the score
- Reach 2 not 3: fraud detection is the payment function. Onboarding and underwriting are named in the same sentence, not shown as separate programs.
- Weight 3 not 2: fraud detection sits in the live payment path.
- Weight 3 not 4: the new AI work in the letter is internal agents. Fraud detection is not a new offer.
- Proof 3 not 4: the warnings are live. The $3.5 billion is potential scam dollars, not a loss rate on the model.
What would move it next quarter: A fraud-loss or false-decline rate tied to the model. Another coding-agent update will not move this.
Open these first
Evidence
Machine learning has run in production from the beginning. Instant seller onboarding, rapid fraud detection, and loan underwriting are described as machine-learning problems solved at scale. The letter's new AI work is internal agents, including code generation, Moneybot, and Managerbot.