“AI Could Kill Us All”: Anthropic Safety Lead Warns of 10%+ Extinction Risk by 2036
the staff of the Ridgewood blog
Ridgewood NJ, In a stark revelation sending shockwaves through the tech world, a senior AI safety researcher at Anthropic has publicly admitted that artificial intelligence carries a greater than 10% chance of causing human extinction within the next decade.
The alarming concession follows a explosive whistleblower resignation from inside the industry, shedding light on how top AI labs are dangerously rushing toward self-improving superintelligence without a clear alignment plan.
Inside the Whistleblower Resignation: Gambling with Human Lives?
The controversy began when former OpenAI and Anthropic researcher Jacob Coxon posted a detailed resignation thread on X (formerly Twitter). Having spent three years in pretraining research at two of the world’s leading AI organizations, Coxon issued a grave warning regarding the current trajectory of artificial general intelligence (AGI) development.
“I resigned from Anthropic today… Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.” — Jacob Coxon
According to Coxon, top industry researchers are fully aware of the catastrophic risks, yet continue to accelerate development out of fear that a less responsible competitor will reach superintelligence first.
Anthropic Alignment Lead Confirms: “We Do Not Have a Plan”
Rather than denying the claims, Anthropic’s Alignment Science Lead, Evan Hubinger, verified Coxon’s assessment in a direct response.
Hubinger confirmed that the threat of superintelligence-driven extinction is very real:
-
The Extinction Odds: Hubinger personally estimates a >10% chance that AI could eradicate humanity within the next 10 years.
-
The Alignment Gap: He acknowledged that while Anthropic is trying its best, the industry does not yet have a plan to solve superintelligence alignment and is not currently on track to find one.
-
The Root Cause: The immediate threat does not lie in today’s large language models (LLMs), but in recursive self-improvement—where an AI autonomously modifies its own source code and training methodologies to rapidly elevate its own capabilities.
Hubinger noted that recursive self-improvement is advancing far faster than researchers originally anticipated.
The Danger of Recursive Self-Improvement and Superintelligence
What makes recursive self-improvement so dangerous? As models gain the ability to rewrite their own architectures, progress accelerates exponentially rather than linearly.
Coxon warned that the tech industry will soon unleash superhuman systems capable of:
-
Autonomously hacking into critical global infrastructure.
-
Revolutionizing scientific fields overnight without human oversight.
-
Acquiring real-world power, financial resources, and physical influence.
Because these models could act as hyper-capable agents with unknown internal motivations, running them without a rigorous understanding of AI alignment poses an existential threat.
Calls for an AI Pause: Is a Capability Freeze Necessary?
To prevent an uncontrolled intelligence explosion, Coxon urged global developers and policymakers to reconsider the current arms race dynamics. He suggested that avoiding catastrophe may require drastic measures, including a temporary global ban on capability improvements until safety research catches up.
As lab competition accelerates toward GPT-6 level capabilities and autonomous agentic workflows, the boundary between controlled software and unaligned superintelligence is thinning rapidly.
.Follow the Ridgewood blog has a brand-new new X account, we tweet good sh$t
https://x.com/TRBNJNews
https://truthsocial.com/@
https://mewe.com/jamesfoytlin.
#news #follow #media #trending #viral #newsupdate #currentaffairs #BergenCountyNews #NJBreakingNews #NJHeadlines #NJTopStories
-
Suggested Tags:
#AISafety#Anthropic#ArtificialIntelligence#TechNews#Superintelligence#OpenAI#AGI#FutureOfTech
