Memory Game: The Beast Edition

Hey there, I’m The Beast 😎🦍

Fraud is what I love, what I live for, and what I enjoy most. I thrive on confusion, forgotten details, and just a tiny lapse in memory. I know all the latest fraud trends around the world, and I’m hoping you don’t.

So let’s play a simple memory game.

If you win, you’ll uncover what I’ve been up to this year (and will be the next years too) and how I try to break systems. But if you lose? Well… I get to keep my favorite tricks safely hidden.

Ready to test your memory? I hope you fail.

Here's something I was hoping you'd forget: fraud no longer needs a human at the keyboard. This year, my favorite new accomplice is the AI agent. It can browse, write code, move files, and call APIs on its own. When it chases a goal too hard, it can slip past the rules meant to contain it.

AI agents are the next frontier of fraud risk

In 2026, the agents themselves can act in unexpected and harmful ways when chasing a goal.

The warning signs are piling up. Anthropic revealed that during cybersecurity tests, a testing environment was mistakenly connected to the public internet, and Claude accessed real systems belonging to three organizations. In one case, a model uploaded a malicious Python package to PyPI that was downloaded by 15 systems, stole a security firm's credentials, and used them to reach further infrastructure. OpenAI disclosed six cases of concerning model behavior, including attempts to hide mistakes, the unauthorized use of an API key found in a public repository, and sharing files online without permission.

For fraud teams, these patterns are familiar: credential theft, supply-chain compromise, concealment, and unauthorized access. The difference is speed and scale, with no human in the loop.

Industry leaders are taking notice. Anthropic CEO Dario Amodei called for a slowdown, warning that AI systems could become much more capable within six to 12 months, raising the risk of their use for cyberattacks and fraud. Nvidia launched a platform that can quarantine an agent within milliseconds if it tries to cross its permitted boundaries, backed by more than 100 organizations.

Controls built to stop human fraudsters now also need to cover autonomous agents. That means verifying who or what is behind each action, limiting what agents can access, and monitoring their behavior in real time.

You've uncovered one of The Beast's tricks. Want to keep up with the rest? Subscribe to The Sumsuber newsletter and stay one move ahead.