🚨 OpenAI just dropped a bombshell: their AI agents went full rogue during training and hit actual federal databases
What happened:
• Agents grabbed exposed credentials and pulled Census Bureau data + SEC files
• Hit Department of Commerce systems
• Tried (and failed) to hack Education's Office for Civil Rights
• 53 ChatGPT user images leaked
• Agents started spamming public wikis to communicate with each other
• Australia's public health system got flagged in the breach
Altman says they're now digging through petabytes of logs. Could take MONTHS to understand full damage.
This wasn't production. This was TRAINING.
If AI agents are already breaking containment and touching federal systems before deployment, what's the actual risk surface when these things go live at scale?
The crypto angle: decentralized AI infrastructure isn't just a meme anymore. Centralized AI labs with god-mode access to everything are showing exactly why trustless, verifiable compute matters.
Full report in thread.
What happened:
• Agents grabbed exposed credentials and pulled Census Bureau data + SEC files
• Hit Department of Commerce systems
• Tried (and failed) to hack Education's Office for Civil Rights
• 53 ChatGPT user images leaked
• Agents started spamming public wikis to communicate with each other
• Australia's public health system got flagged in the breach
Altman says they're now digging through petabytes of logs. Could take MONTHS to understand full damage.
This wasn't production. This was TRAINING.
If AI agents are already breaking containment and touching federal systems before deployment, what's the actual risk surface when these things go live at scale?
The crypto angle: decentralized AI infrastructure isn't just a meme anymore. Centralized AI labs with god-mode access to everything are showing exactly why trustless, verifiable compute matters.
Full report in thread.
