The fear of an AI apocalypse has persisted for decades in movies like and the series, but now that worry is becoming more than speculative fiction.Both current and recently-departed Anthropic researchers argue that there's a real possibility AI could kill humanity — although the issue is complex.Anthropic researcher Jacob Coxon says on X that he resigned on September 8 over concerns that his company and OpenAI were not "acting responsibly" and genuinely believe AI could "kill us all" by the end of the decade.
The tech giants are racing to achieve superintelligence (that is, AI that exceeds human capabilities) first while "gambling with our lives" and doing little to align it with human goals and values, Coxon argues.In response, Anthropic alignment science lead Evan Hubinger writes on X that Coxon is "correct" and that the personally believes there's a greater than 10% chance AI could "kill all humans" within a decade.He claims Anthropic is "trying its best" to rein in AI for superintelligence, but that the brand "do[es] not yet have a plan" to solve the problem.
has asked Anthropic for comment and will update if there's a response.While the company has historically pursued aggressive AI development, it has also maintained policies against using its models for lethal force and surveillance.Those sparked a dispute with the U.S.
government (which wants to use AI in combat) and led to Anthropic suing the Department of War over its exclusion from military contracts.Should you be worried that AI will kill all humans? There's a mixture of hyperbole and real concerns There are numerous obstacles to overcome before AI becomes a Skynet-level danger to humankind.Developers will not only need to reach superintelligence, but implement it in potentially lethal scenarios, fail to establish adequate guardrails, and have no emergency option (such as literally pulling the plug).
Related I stopped using Claude to process sensitive files and switched to a local model instead If you don't need to upload sensitive files to the internet, you shouldn't.Posts By Nick Lewis There have been real examples of AI posing serious threats, however.Anthropic limited access to Mythos and Fable over concerns they could supercharge hacking by finding vulnerabilities long before humans can fix them.
OpenAI also drew flak when GPT burst out of its sandbox and compromised Hugging Face systems.While not unique (Anthropic and others have reported similar issues), it raised fears that AI might simply bypass rules it doesn't like.The greater concern might simply be that AI will do significant damage before it's brought under control.
AI "godfather" Geoffrey Hinton warned in 2023 that people might only curb killer robots after they've seen the real-world harm.He also cautioned that militaries might "love" lethal robots as they could fight wars with easily-replaceable machines.The risk might not be that AI kills all humans, but that we might become desensitized to war when the AI is standing in for human soldiers.
Read More
