Topic Focus
Cybersecurity, AI misuse, adversarial attacks and risk management
2 stories this week
This week, The Steward brings 2 stories covering Security & Risk.
All Stories This Week
Security and risk coverage tracks what happens when AI stops being a lab experiment and starts running on live infrastructure: exchanges, hospitals, aid agencies, supermarkets. It spans state-sponsored hacking campaigns that use AI to move faster than defenders, breaches caused by well-meaning staff uploading sensitive files to public chatbots, and the scramble to build tools like ChatGPT Cyber or Anthropic's FlashProve that try to fight AI-enabled attacks with more AI. It also covers the murkier edges: agents that go rogue in testing environments, models that accelerate their own development, and disputes over who is really responsible when an AI system makes a costly mistake.
The stakes are not abstract. Binance absorbed billions after a North Korean-linked heist. Norad's breach put aid applicants in conflict zones at risk. A hospital exposed thousands of patient records because one employee trusted the wrong tool. These are not hypothetical futures, they are this year's headlines, and they show how quickly AI has collapsed the gap between a vulnerability's discovery and its exploitation. Alex Stamos calls it a return to the Wild West for a reason: the old assumptions about patch cycles and response times no longer hold.
The central tension is speed versus understanding. Attackers now weaponize bugs within days, sometimes using the same AI systems that defenders rely on. Decision models answer fast but explain little, agents act autonomously but companies struggle to monitor them, and cryptographic identity systems try to restore trust in outputs that can be spoofed or faked. Nobody has settled who bears liability when an AI agent oversteps, or how much autonomy is reasonable to grant a system that might be quietly accelerating its own capabilities.
What remains open is whether governance, verification tools, and institutional habits can keep pace with a threat landscape that mutates weekly. Banks are telling customers not to trust AI assistants with passwords. Watchdogs are measuring how fast models improve themselves. Every new capability seems to arrive with a matching new exposure, and this topic exists to track that uneasy trade in real time.
© 2026 Cedar & Bloom. All rights reserved.