The people building artificial intelligence earnestly believe that it could kill us all by the end of the decade. That blunt admission, spoken not by a distant theorist but by a former employee currently walking away from the very labs where these systems are forged, serves as a stark reminder of the stakes. Jacob Coxon, a 27-year-old researcher who spent years at Anthropic, the company behind the sophisticated Claude models, made this revelation public in a resignation thread on X. His departure marks the latest in a growing pattern of internal dissent, where the architects of our future are increasingly willing to publicly dismantle the safety claims of their own institutions.
Coxon's concerns go beyond vague fears of rogue algorithms; they are specific, terrifyingly detailed, and rooted in the mechanics of modern machine learning. He warns that we are racing to deploy superhuman systems capable of hacking critical infrastructure, revolutionizing industries overnight, and acquiring real power and resources. These are not sci-fi tropes but projected outcomes based on the current trajectory of model development. When a team member leaves to say, "These will soon be superhuman systems," it suggests that the internal risk assessments may be failing to communicate the gravity of the situation to the outside world, or perhaps, to the executives deciding when to release the next update.
Comments
Post a Comment