Jon Swartz, Chris Blask, Andi Mann and Teri Robinson break down three stories that show how quickly AI is colliding with real-world risk: a Cursor/Claude-powered coding agent tied to a Railway production data-loss scare, the Trump administration’s push to use AI to modernize air traffic planning, and the FBI-led takedown of the W3LL phishing-kit ecosystem tied to more than $20 million in attempted fraud.

The first segment looks at what happens when agentic coding tools get too much power in production. Reports say the PocketOS incident involved deletion of a production database and recent backups after an AI agent used a destructive Railway API path; Railway later recovered service quickly and patched the endpoint, but the episode exposed major questions around permissions, rollback design and human oversight.

The second segment turns to aviation. The FAA’s broader modernization push includes new software intended to improve airspace management and reduce disruption, while SMART-style planning tools are being positioned as a move from reactive to predictive operations. That raises the obvious question: where should AI sit inside a safety-critical system, and who owns accountability when predictive systems shape operational decisions?

The final segment focuses on phishing at industrial scale. The FBI says the W3LL phishing kit was sold for about $500, helped criminals impersonate trusted login pages, captured credentials and session data to bypass MFA, and was tied to over $20 million in attempted fraud; investigators also say W3LLSTORE facilitated sales of more than 25,000 compromised accounts from 2019 to 2023.

#TechstrongGang #AI #Cybersecurity #AIAgents #Phishing