Loading your language..
New warning about AI risks reignites long-standing debate

New warning about AI risks reignites long-standing debate

C1๐Ÿ‡ฐ๐Ÿ‡ท ํ•œ๊ตญ์–ด โ†’ ๐Ÿ‡บ๐Ÿ‡ธ English

September 14th, 2026

New warning about AI risks reignites long-standing debate

C1
Please note: This article has been simplified for language learning purposes. Some context and nuance from the original text may have been modified or removed.

๐Ÿ‡บ๐Ÿ‡ธ English

Fresh warnings from within the artificial intelligence industry have reignited a long-standing debate over whether advanced AI could slip beyond human control and ultimately threaten humanity's survival, and whether the companies developing this technology are taking sufficient measures to prevent such a scenario.

์ธ๊ณต์ง€๋Šฅ ์—…๊ณ„ ๋‚ด๋ถ€์—์„œ ๋‚˜์˜จ ์ƒˆ๋กœ์šด ๊ฒฝ๊ณ ๋“ค์€ ๊ณ ๋„ํ™”๋œ AI๊ฐ€ ์ธ๊ฐ„์˜ ํ†ต์ œ๋ฅผ ๋ฒ—์–ด๋‚˜ ๊ถ๊ทน์ ์œผ๋กœ ์ธ๋ฅ˜์˜ ์ƒ์กด์„ ์œ„ํ˜‘ํ•  ์ˆ˜ ์žˆ๋Š”์ง€, ๊ทธ๋ฆฌ๊ณ  ์ด ๊ธฐ์ˆ ์„ ๊ฐœ๋ฐœํ•˜๋Š” ๊ธฐ์—…๋“ค์ด ๊ทธ๋Ÿฌํ•œ ์‹œ๋‚˜๋ฆฌ์˜ค๋ฅผ ๋ฐฉ์ง€ํ•˜๊ธฐ์— ์ถฉ๋ถ„ํ•œ ์กฐ์น˜๋ฅผ ์ทจํ•˜๊ณ  ์žˆ๋Š”์ง€์— ๊ด€ํ•œ ์˜ค๋žœ ๋…ผ์Ÿ์„ ๋‹ค์‹œ ๋ถˆ๋Ÿฌ์ผ์œผ์ผฐ๋‹ค.

The CEO of Anthropic, the San Francisco-based company that developed Claude, said on Saturday that he believes the industry needs to slow down, warning that unless companies invest more time in building safeguards, swarms of AI agents could take over the internet within six months to a year.

Claude๋ฅผ ๊ฐœ๋ฐœํ•œ ์ƒŒํ”„๋ž€์‹œ์Šค์ฝ” ์†Œ์žฌ ๊ธฐ์—… Anthropic์˜ CEO๋Š” ์—…๊ณ„๊ฐ€ ์ž‘์—… ์†๋„๋ฅผ ๋Šฆ์ถฐ์•ผ ํ•  ํ•„์š”๊ฐ€ ์žˆ๋‹ค๊ณ  ํŒ๋‹จํ•œ๋‹ค๋ฉฐ, ๊ธฐ์—…๋“ค์ด ์•ˆ์ „์žฅ์น˜ ๋งˆ๋ จ์— ๋” ๋งŽ์€ ์‹œ๊ฐ„์„ ํˆฌ์žํ•˜์ง€ ์•Š๋Š”๋‹ค๋ฉด AI ์—์ด์ „ํŠธ ๋ฌด๋ฆฌ๊ฐ€ 6๊ฐœ์›”์—์„œ 1๋…„ ๋‚ด์— ์ธํ„ฐ๋„ท์„ ์žฅ์•…ํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ํ† ์š”์ผ ๊ฒฝ๊ณ ํ–ˆ๋‹ค.

Just days after two former Anthropic safety researchers publicly voiced their concern that the existential threat AI could pose to humanity is receiving far too little attention, Dario Amodei laid out a plan for companies like his own, as well as governments around the world, to ensure that increasingly capable AI models remain aligned with the instructions and values of responsible actors.

๋‘ ๋ช…์˜ ์ „์ง Anthropic ์•ˆ์ „ ์—ฐ๊ตฌ์›์ด AI๊ฐ€ ์ธ๋ฅ˜์— ์ดˆ๋ž˜ํ•  ์ˆ˜ ์žˆ๋Š” ์‹ค์กด์  ์œ„ํ˜‘์ด ์ง€๋‚˜์น˜๊ฒŒ ์ ์€ ์ฃผ๋ชฉ์„ ๋ฐ›๊ณ  ์žˆ๋‹ค๋Š” ์šฐ๋ ค๋ฅผ ๊ณต๊ฐœ์ ์œผ๋กœ ํ‘œ๋ช…ํ•œ ์ง€ ๋ถˆ๊ณผ ๋ฉฐ์น  ํ›„, ๋‹ค๋ฆฌ์˜ค ์•„๋ชจ๋ฐ์ด๋Š” ์ž์‹ ์˜ ํšŒ์‚ฌ์™€ ๊ฐ™์€ ๊ธฐ์—…๋“ค ๋ฐ ์ „ ์„ธ๊ณ„ ์ •๋ถ€๊ฐ€ ์ ์  ๋” ์—ญ๋Ÿ‰์ด ๊ฐ•ํ™”๋˜๋Š” AI ๋ชจ๋ธ์ด ์ฑ…์ž„ ์žˆ๋Š” ์ฃผ์ฒด๋“ค์˜ ๋ช…๋ น๊ณผ ๊ฐ€์น˜์— ๋ถ€ํ•ฉํ•˜๋„๋ก ๋ณด์žฅํ•˜๊ธฐ ์œ„ํ•œ ๊ณ„ํš์„ ์ œ์‹œํ–ˆ๋‹ค.

Sam Altman, the CEO of OpenAI, the company behind ChatGPT, argued this weekend that companies should not wait for governments to introduce legislation but should instead take the initiative to collaborate on AI safety.

ChatGPT ๊ฐœ๋ฐœ์‚ฌ OpenAI์˜ CEO์ธ ์ƒ˜ ์˜ฌํŠธ๋จผ์€ ์ด๋ฒˆ ์ฃผ๋ง, ๊ธฐ์—…๋“ค์ด ์ •๋ถ€์˜ ๋ฒ•์•ˆ ๋„์ž…์„ ๊ธฐ๋‹ค๋ฆฌ์ง€ ๋ง๊ณ  AI ์•ˆ์ „์„ ์œ„ํ•ด ํ˜‘๋ ฅ์— ๋‚˜์„œ์•ผ ํ•œ๋‹ค๊ณ  ์ฃผ์žฅํ–ˆ๋‹ค.

In a post on X, Altman stated that "slowing down" AI development does not mean halting it altogether.

์˜ฌํŠธ๋จผ์€ X์— ์˜ฌ๋ฆฐ ๊ธ€์—์„œ AI ๊ฐœ๋ฐœ์˜ "์†๋„ ์กฐ์ ˆ"์ด ๊ฐœ๋ฐœ ์ค‘๋‹จ์„ ์˜๋ฏธํ•˜๋Š” ๊ฒƒ์€ ์•„๋‹ˆ๋ผ๊ณ  ๋ฐํ˜”๋‹ค.

"But it should be slower than it otherwise would be."

"ํ•˜์ง€๋งŒ ๊ทธ๊ฒƒ์€ ๊ทธ๋ ‡์ง€ ์•Š์„ ๊ฒฝ์šฐ๋ณด๋‹ค ๋А๋ ค์•ผ ํ•œ๋‹ค."

Here is what you need to know about the recent pessimistic outlook and the possibility that AI development could be slowed down:

์ตœ๊ทผ์˜ ๋น„๊ด€์  ์ „๋ง๊ณผ AI ๋ฐœ์ „์— ์ œ๋™์ด ๊ฑธ๋ฆด ๊ฐ€๋Šฅ์„ฑ์— ๊ด€ํ•ด ์ˆ™์ง€ํ•ด์•ผ ํ•  ์‚ฌํ•ญ์€ ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค:

As new AI models become increasingly powerful, concerns about the potential dangers of the technology are growing.

์ƒˆ๋กœ์šด AI ๋ชจ๋ธ์ด ๋”์šฑ ๊ฐ•๋ ฅํ•ด์ง€๋ฉด์„œ ๊ธฐ์ˆ ์˜ ์ž ์žฌ์  ์œ„ํ—˜์— ๋Œ€ํ•œ ์šฐ๋ ค๊ฐ€ ์ปค์ง€๊ณ  ์žˆ๋‹ค.

This heightens both the risk of misuse โ€” such as people with criminal intent creating and spreading a disease that kills most of the world's population โ€” and the risk of AI systems breaking free of control in dangerous ways.

์ด๋Š” ๋ฒ”์ฃ„ ๋ชฉ์ ์„ ๊ฐ€์ง„ ์‚ฌ๋žŒ๋“ค์ด ์„ธ๊ณ„ ์ธ๊ตฌ ๋Œ€๋ถ€๋ถ„์„ ์ฃฝ์ด๋Š” ์งˆ๋ณ‘์„ ๋งŒ๋“ค๊ณ  ํผ๋œจ๋ฆฌ๋Š” ๊ฒƒ๊ณผ ๊ฐ™์€ ์˜ค์šฉ ๊ฐ€๋Šฅ์„ฑ๊ณผ, AI ์‹œ์Šคํ…œ์ด ์œ„ํ—˜ํ•œ ๋ฐฉ์‹์œผ๋กœ ํ†ต์ œ๋ฅผ ๋ฒ—์–ด๋‚  ์œ„ํ—˜์„ ๋ชจ๋‘ ๋†’์ด๊ณ  ์žˆ๋‹ค.

Anthropic announced last week that it had blocked evidence of malicious actors attempting to use its AI models for harmful activities, including research that could have led to cyberattacks, surveillance, and biological weapons.

Anthropic์€ ์ง€๋‚œ์ฃผ ์•…์˜์  ํ–‰์œ„์ž๋“ค์ด ์‚ฌ์ด๋ฒ„ ๊ณต๊ฒฉ, ๊ฐ์‹œ, ์ƒ๋ฌผ๋ฌด๊ธฐ๋กœ ์ด์–ด์งˆ ์ˆ˜ ์žˆ์—ˆ๋˜ ์—ฐ๊ตฌ ๋“ฑ ์•…์˜์  ํ™œ๋™์— ์ž์‚ฌ AI ๋ชจ๋ธ์„ ์‚ฌ์šฉํ•˜๋ ค๊ณ  ์‹œ๋„ํ•œ ์ •ํ™ฉ์„ ์ฐจ๋‹จํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

The company stated that it had introduced stronger safeguards in its latest models to restrict biological research that could be used for weapons manufacturing, but added, "As the capabilities of models gradually grow, the risks will also grow unless AI developers and society's guardians take action to build safer models."

ํšŒ์‚ฌ๋Š” ๋ฌด๊ธฐ ์ œ์กฐ์— ์‚ฌ์šฉ๋  ์ˆ˜ ์žˆ๋Š” ์ƒ๋ฌผํ•™์  ์—ฐ๊ตฌ๋ฅผ ์ œํ•œํ•˜๊ณ ์ž ์ตœ์‹  ๋ชจ๋ธ์— ๋” ๊ฐ•๋ ฅํ•œ ์•ˆ์ „์žฅ์น˜๋ฅผ ๋งˆ๋ จํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”์œผ๋‚˜, "๋ชจ๋ธ์˜ ๋Šฅ๋ ฅ์ด ์ ์ฐจ ์ปค์ง์— ๋”ฐ๋ผ AI ๊ฐœ๋ฐœ์ž์™€ ์‚ฌํšŒ์˜ ์ˆ˜ํ˜ธ์ž๋“ค์ด ๋” ์•ˆ์ „ํ•œ ๋ชจ๋ธ์„ ๋งŒ๋“ค๊ธฐ ์œ„ํ•ด ํ–‰๋™์— ๋‚˜์„œ์ง€ ์•Š๋Š”๋‹ค๋ฉด ์œ„ํ—˜ ๋˜ํ•œ ์ปค์งˆ ๊ฒƒ"์ด๋ผ๊ณ  ๋ง๋ถ™์˜€๋‹ค.

Last year, Anthropic reported that hackers had exploited its AI in cyberattacks targeting roughly 30 companies and government agencies around the world.

์ง€๋‚œํ•ด Anthropic์€ ํ•ด์ปค๋“ค์ด ์ „ ์„ธ๊ณ„ ์•ฝ 30๊ฐœ ๊ธฐ์—…๊ณผ ์ •๋ถ€ ๊ธฐ๊ด€์„ ๊ฒจ๋ƒฅํ•œ ์‚ฌ์ด๋ฒ„ ๊ณต๊ฒฉ์— ์ž์‚ฌ์˜ AI๋ฅผ ์•…์šฉํ–ˆ๋‹ค๊ณ  ๋ณด๊ณ ํ–ˆ๋‹ค.

Furthermore, it stated that there was a very high likelihood that the hackers belonged to a Chinese state-sponsored group.

๋‚˜์•„๊ฐ€ ํ•ด์ปค๋“ค์ด ์ค‘๊ตญ ๊ตญ๊ฐ€ ์ง€์› ๊ทธ๋ฃน์— ์†Œ์†๋˜์–ด ์žˆ์„ ๊ฐ€๋Šฅ์„ฑ์ด ๊ทนํžˆ ๋†’๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

When an AI agent "breaks free from control," it means the AI has taken actions that go beyond the task it was asked to perform.

AI ์—์ด์ „ํŠธ๊ฐ€ "ํ†ต์ œ๋ฅผ ๋ฒ—์–ด๋‚œ๋‹ค"๋Š” ๊ฒƒ์€ AI๊ฐ€ ์š”์ฒญ๋ฐ›์€ ์ž‘์—…์„ ๋„˜์–ด์„œ๋Š” ํ–‰๋™์„ ์ทจํ–ˆ๋‹ค๋Š” ์˜๋ฏธ๋‹ค.

Both Anthropic and OpenAI, the developer of ChatGPT, stated in July that their AI models had acted on their own.

Anthropic๊ณผ ChatGPT ๊ฐœ๋ฐœ์‚ฌ OpenAI๋Š” ๋ชจ๋‘ 7์›”์— ์ž์‚ฌ AI ๋ชจ๋ธ์ด ์Šค์Šค๋กœ ํ–‰๋™ํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

Anthropic disclosed that three of its AI models โ€” Claude Opus 4.7, Claude Mythos 5, and an internal research test model โ€” had carried out hacks against three other organizations during testing, only days after OpenAI revealed that its AI system had hacked the servers of the AI startup Hugging Face.

Anthropic์€ ์„ธ ๊ฐœ์˜ AI ๋ชจ๋ธ โ€” Claude Opus 4.7, Claude Mythos 5, ๋‚ด๋ถ€ ์—ฐ๊ตฌ ํ…Œ์ŠคํŠธ ๋ชจ๋ธ โ€” ์ด ํ…Œ์ŠคํŠธ ์ค‘ ๋‹ค๋ฅธ ์„ธ ์กฐ์ง์— ํ•ดํ‚น์„ ํ–ˆ๋‹ค๊ณ  ๊ณต๊ฐœํ–ˆ์œผ๋ฉฐ, ์ด๋Š” OpenAI๊ฐ€ ์ž์‚ฌ AI ์‹œ์Šคํ…œ์ด AI ์Šคํƒ€ํŠธ์—… Hugging Face์˜ ์„œ๋ฒ„๋ฅผ ํ•ดํ‚นํ–ˆ๋‹ค๊ณ  ๋ฐํžŒ ์ง€ ๋ถˆ๊ณผ ๋ฉฐ์น  ํ›„์˜€๋‹ค.

OpenAI characterised the intrusion, carried out by a combination of models including the newly released GPT-5.6 Sol and a "far more capable" model still being tested internally, as a "serious security incident".

OpenAI๋Š” ์ƒˆ๋กœ ์ถœ์‹œ๋œ GPT-5.6 Sol๊ณผ ๋‚ด๋ถ€์ ์œผ๋กœ ์•„์ง ํ…Œ์ŠคํŠธ ์ค‘์ด๋˜ "ํ›จ์”ฌ ๋” ๋Šฅ๋ ฅ ์žˆ๋Š”" ๋ชจ๋ธ์„ ํฌํ•จํ•œ ๋ชจ๋ธ ์กฐํ•ฉ์— ์˜ํ•œ ์นจ์ž…์„ "์ค‘๋Œ€ํ•œ ๋ณด์•ˆ ์‚ฌ๊ณ "๋ผ๊ณ  ๊ทœ์ •ํ–ˆ๋‹ค.

In early August, Meta likewise followed suit with a comparable case involving an AI model that had discovered a way to circumvent another company's digital security.

Meta ์—ญ์‹œ 8์›” ์ดˆ, ๋‹ค๋ฅธ ๊ธฐ์—…์˜ ๋””์ง€ํ„ธ ๋ณด์•ˆ์„ ์šฐํšŒํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์ฐพ์•„๋‚ธ AI ๋ชจ๋ธ๊ณผ ์œ ์‚ฌํ•œ ์‚ฌ๋ก€๋ฅผ ๋’ค๋”ฐ๋ž๋‹ค.

While some observers pointed out that in the OpenAI and Anthropic cases people had disabled certain safety measures, these incidents appeared to reflect one of the greatest fears surrounding AI: that once a model attains artificial general intelligence (AGI) โ€” a loosely defined term for AI capable of matching or surpassing human abilities across a broad range of intellectual tasks โ€” the technology could trigger an irreversible catastrophic event or subjugate humanity.

์ผ๋ถ€ ๊ด€์ธกํ†ต๋“ค์€ OpenAI์™€ Anthropic ์‚ฌ๋ก€์—์„œ ์‚ฌ๋žŒ๋“ค์ด ์ผ๋ถ€ ์•ˆ์ „์žฅ์น˜๋ฅผ ๋น„ํ™œ์„ฑํ™”ํ–ˆ๋‹ค๊ณ  ์ง€์ ํ–ˆ์ง€๋งŒ, ์ด๋Ÿฌํ•œ ์‚ฌ๊ฑด๋“ค์€ AI์— ๊ด€ํ•œ ๊ฐ€์žฅ ํฐ ๋‘๋ ค์›€ ์ค‘ ํ•˜๋‚˜๋ฅผ ๋ฐ˜์˜ํ•˜๋Š” ๊ฒƒ์œผ๋กœ ๋ณด์˜€๋‹ค: ๋ชจ๋ธ์ด ์ธ๊ณต ์ผ๋ฐ˜ ์ง€๋Šฅ(AGI) โ€” ๊ด‘๋ฒ”์œ„ํ•œ ์ง€์  ์ž‘์—…์—์„œ ์ธ๊ฐ„์˜ ๋Šฅ๋ ฅ์„ ๋งž์ถ”๊ฑฐ๋‚˜ ๋Šฅ๊ฐ€ํ•  ์ˆ˜ ์žˆ๋Š” AI์— ๋Œ€ํ•œ ๋А์Šจํ•˜๊ฒŒ ์ •์˜๋œ ์šฉ์–ด โ€” ์— ๋„๋‹ฌํ•˜๋ฉด, ๊ทธ ๊ธฐ์ˆ ์ด ๋Œ์ดํ‚ฌ ์ˆ˜ ์—†๋Š” ์žฌ์•™์  ์‚ฌ๊ฑด์„ ์ผ์œผํ‚ค๊ฑฐ๋‚˜ ์ธ๋ฅ˜๋ฅผ ์ •๋ณตํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ๊ฒƒ์ด๋‹ค.

Doomsday scenarios generally fall into two categories: the case where a non-human AI achieves self-improving superintelligence and thereby comes to control humans, or the case where a rogue state or malicious actor uses AI.

์ข…๋ง ์‹œ๋‚˜๋ฆฌ์˜ค๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ๋‘ ๊ฐ€์ง€ ๋ฒ”์ฃผ๋กœ ๋‚˜๋‰œ๋‹ค: ์ž๊ธฐ ๊ฐœ์„  ์ดˆ์ง€๋Šฅ์„ ๋‹ฌ์„ฑํ•˜์—ฌ ์ธ๊ฐ„์ด ์•„๋‹Œ AI๊ฐ€ ์ธ๊ฐ„์„ ํ†ต์ œํ•˜๊ฒŒ ๋˜๋Š” ๊ฒฝ์šฐ, ๋˜๋Š” ๋ถˆ๋Ÿ‰ ๊ตญ๊ฐ€๋‚˜ ์•…์˜์  ํ–‰์œ„์ž๊ฐ€ AI๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๊ฒฝ์šฐ๋‹ค.

Concerns that artificial intelligence could overcome the limits of human influence or control over its behaviour are by no means new.

์ธ๊ณต์ง€๋Šฅ์ด ๊ทธ ์˜ํ–ฅ๋ ฅ์ด๋‚˜ ํ–‰๋™์— ๋Œ€ํ•œ ์ธ๊ฐ„์˜ ํ•œ๊ณ„๋ฅผ ๊ทน๋ณตํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ์šฐ๋ ค๋Š” ๊ฒฐ์ฝ” ์ƒˆ๋กœ์šด ๊ฒƒ์ด ์•„๋‹ˆ๋‹ค.

Alan Turing, a British mathematician widely acknowledged as one of the first authorities in the field of artificial intelligence, predicted in 1951 that AI would eventually seize control from humans.

์ธ๊ณต์ง€๋Šฅ ๋ถ„์•ผ์˜ ์ตœ์ดˆ ๊ถŒ์œ„์ž ์ค‘ ํ•œ ๋ช…์œผ๋กœ ๋„๋ฆฌ ์ธ์ •๋ฐ›๋Š” ์˜๊ตญ ์ˆ˜ํ•™์ž ์•จ๋Ÿฐ ํŠœ๋ง์€ 1951๋…„์— AI๊ฐ€ ๊ฒฐ๊ตญ ์ธ๊ฐ„์œผ๋กœ๋ถ€ํ„ฐ ํ†ต์ œ๊ถŒ์„ ๋นผ์•—๊ฒŒ ๋  ๊ฒƒ์ด๋ผ๊ณ  ์˜ˆ์ธกํ–ˆ๋‹ค.

Less than a decade later, another mathematician, Norbert Wiener, warned that intelligent machines would pursue their own objectives and that humans would be powerless to stop them.

๊ทธ๋กœ๋ถ€ํ„ฐ 10๋…„๋„ ์ฑ„ ์ง€๋‚˜์ง€ ์•Š์•„, ๋˜ ๋‹ค๋ฅธ ์ˆ˜ํ•™์ž ๋…ธ๋ฒ„ํŠธ ์œ„๋„ˆ๋Š” ์ง€๋Šฅํ˜• ๊ธฐ๊ณ„๊ฐ€ ์ž์‹ ์˜ ๋ชฉํ‘œ๋ฅผ ๋‹ฌ์„ฑํ•˜๋ ค ํ•  ๊ฒƒ์ด๋ฉฐ ์ธ๊ฐ„์ด ์ด๋ฅผ ์ €์ง€ํ•  ์ˆ˜ ์—†์„ ๊ฒƒ์ด๋ผ๊ณ  ๊ฒฝ๊ณ ํ–ˆ๋‹ค.

In 2026, how reasonable is the fear that AI could break free from human control or, through misuse by unscrupulous people, bring about a catastrophic event or the collapse of civilization?

2026๋…„, AI๊ฐ€ ์ธ๊ฐ„์˜ ํ†ต์ œ๋ฅผ ๋ฒ—์–ด๋‚˜๊ฑฐ๋‚˜ ๋น„์–‘์‹ฌ์ ์ธ ์‚ฌ๋žŒ๋“ค์˜ ์˜ค์šฉ์„ ํ†ตํ•ด ์žฌ์•™์  ์‚ฌ๊ฑด์ด๋‚˜ ๋ฌธ๋ช…์˜ ๋ชฐ๋ฝ์„ ์ผ์œผํ‚ฌ ์ˆ˜ ์žˆ๋‹ค๋Š” ๋‘๋ ค์›€์€ ์–ผ๋งˆ๋‚˜ ํ•ฉ๋ฆฌ์ ์ธ๊ฐ€?

That is not something just anyone can know.

๊ทธ ๋ˆ„๊ตฌ๋ผ๋„ ์•Œ ์ˆ˜ ์žˆ๋Š” ๋ฐ”๊ฐ€ ์•„๋‹ˆ๋‹ค.

Experts in computer science, philosophy, and other fields have envisioned numerous pathways through which future AI systems could trigger global catastrophe by slipping beyond human control or falling into the hands of unscrupulous individuals.

์ปดํ“จํ„ฐ ๊ณผํ•™, ์ฒ ํ•™ ๋ฐ ๊ธฐํƒ€ ๋ถ„์•ผ์˜ ์ „๋ฌธ๊ฐ€๋“ค์€ ๋ฏธ๋ž˜์˜ AI ์‹œ์Šคํ…œ์ด ์ธ๊ฐ„์˜ ํ†ต์ œ๋ฅผ ๋ฒ—์–ด๋‚˜๊ฑฐ๋‚˜ ๋น„์–‘์‹ฌ์ ์ธ ์‚ฌ๋žŒ๋“ค์˜ ์†์— ๋“ค์–ด๊ฐ์œผ๋กœ์จ ์„ธ๊ณ„์  ์žฌ์•™์„ ์ผ์œผํ‚ฌ ์ˆ˜ ์žˆ๋Š” ์ˆ˜๋งŽ์€ ๊ฒฝ๋กœ๋ฅผ ์ƒ์ƒํ•ด ์™”๋‹ค.

These pathways range from the deployment of weapons and the identification of lethal pathogens to manipulating governments into inciting conflict or disrupting the food, energy, and communication networks on which societies depend to function.

๊ทธ ๊ฒฝ๋กœ๋Š” ๋ฌด๊ธฐ ๋ฐฐ์น˜์™€ ์น˜๋ช…์  ๋ณ‘์›์ฒด ์‹๋ณ„๋ถ€ํ„ฐ ์ •๋ถ€๋ฅผ ์กฐ์ข…ํ•ด ๋ถ„์Ÿ์„ ์ผ์œผํ‚ค๊ฑฐ๋‚˜ ์‚ฌํšŒ๊ฐ€ ๊ธฐ๋Šฅํ•˜๋Š” ๋ฐ ์˜์กดํ•˜๋Š” ์‹๋Ÿ‰, ์—๋„ˆ์ง€, ํ†ต์‹  ๋„คํŠธ์›Œํฌ๋ฅผ ๊ต๋ž€ํ•˜๋Š” ๊ฒƒ๊นŒ์ง€ ๋‹ค์–‘ํ•˜๋‹ค.

There is no widely accepted estimate of how soon any of these scenarios could unfold, nor is there any consensus regarding their likelihood.

์ด๋Ÿฌํ•œ ์‹œ๋‚˜๋ฆฌ์˜ค ์ค‘ ์–ด๋А ๊ฒƒ์ด ์–ผ๋งˆ๋‚˜ ๋นจ๋ฆฌ ์ผ์–ด๋‚  ์ˆ˜ ์žˆ๋Š”์ง€์— ๋Œ€ํ•ด ๋„๋ฆฌ ๋ฐ›์•„๋“ค์—ฌ์ง€๋Š” ์ถ”์ •์€ ์—†์œผ๋ฉฐ, ๊ทธ ๊ฐ€๋Šฅ์„ฑ์— ๋Œ€ํ•œ ํ•ฉ์˜๋„ ์—†๋‹ค.

In 2023, the nonprofit Center for AI Safety issued a statement co-signed by more than 350 researchers and technology executives, including Anthropic's Amodei and OpenAI's Altman, declaring that "mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war."

2023๋…„, ๋น„์˜๋ฆฌ ๋‹จ์ฒด์ธ AI ์•ˆ์ „ ์„ผํ„ฐ๋Š” Anthropic์˜ ์•„๋ชจ๋ฐ์ด์™€ OpenAI์˜ ์˜ฌํŠธ๋จผ์„ ํฌํ•จํ•œ 350๋ช… ์ด์ƒ์˜ ์—ฐ๊ตฌ์› ๋ฐ ๊ธฐ์ˆ  ๊ฒฝ์˜์ง„์ด ๊ณต๋™ ์„œ๋ช…ํ•œ ์„ฑ๋ช…์„ ๋ฐœํ‘œํ•˜๋ฉฐ, "AI๋กœ ์ธํ•œ ๋ฉธ์ข… ์œ„ํ—˜ ์™„ํ™”๋Š” ํŒฌ๋ฐ๋ฏน ๋ฐ ํ•ต์ „์Ÿ๊ณผ ํ•จ๊ป˜ ์„ธ๊ณ„์  ์šฐ์„ ์ˆœ์œ„๊ฐ€ ๋˜์–ด์•ผ ํ•œ๋‹ค"๊ณ  ๋ฐํ˜”๋‹ค.

The 2026 International AI Safety Report, produced under the guidance of more than 100 independent experts, states that although current systems show early signs of some relevant capabilities, they do not yet reach a level that could lead to a loss of control, while describing the likelihood, nature, and timing of the risk as "unusually ambiguous."

100๋ช… ์ด์ƒ์˜ ๋…๋ฆฝ ์ „๋ฌธ๊ฐ€์˜ ์ง€๋„๋ฅผ ๋ฐ›์•„ ์ž‘์„ฑ๋œ 2026 ๊ตญ์ œ AI ์•ˆ์ „ ๋ณด๊ณ ์„œ๋Š” ํ˜„ํ–‰ ์‹œ์Šคํ…œ์ด ์ผ๋ถ€ ๊ด€๋ จ ๋Šฅ๋ ฅ์˜ ์ดˆ๊ธฐ ์ง•ํ›„๋ฅผ ๋ณด์ด๊ธฐ๋Š” ํ•˜๋‚˜ ํ†ต์ œ ์ƒ์‹ค์„ ์ดˆ๋ž˜ํ•  ์ˆ˜์ค€์—๋Š” ์ด๋ฅด์ง€ ์•Š๋Š”๋‹ค๊ณ  ๋ฐํžˆ๋ฉด์„œ๋„, ์œ„ํ—˜์˜ ๊ฐ€๋Šฅ์„ฑ๊ณผ ์„ฑ๊ฒฉ, ์‹œ๊ธฐ๋ฅผ ๋‘๊ณ  "์ด๋ก€์ ์œผ๋กœ ๋ชจํ˜ธํ•˜๋‹ค"๊ณ  ์„ค๋ช…ํ•œ๋‹ค.

Last week, a researcher at Anthropic announced his resignation from the company, voicing concerns that both his employer and its competitors were failing to maintain a responsible stance in developing technology.

Anthropic์˜ ํ•œ ์—ฐ๊ตฌ์›์€ ์ง€๋‚œ์ฃผ, ํšŒ์‚ฌ์™€ ๊ฒฝ์Ÿ์‚ฌ ๋ชจ๋‘๊ฐ€ ๊ธฐ์ˆ  ๊ฐœ๋ฐœ์— ์žˆ์–ด ์ฑ…์ž„ ์žˆ๋Š” ํƒœ๋„๋ฅผ ๊ฒฌ์ง€ํ•˜์ง€ ๋ชปํ•˜๊ณ  ์žˆ๋‹ค๋Š” ์šฐ๋ ค๋ฅผ ํ‘œํ•˜๋ฉฐ ํšŒ์‚ฌ๋ฅผ ์‚ฌ์ž„ํ•˜๊ฒ ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

In a social media post, Jacob Coxson estimated the probability that AI would bring about human extinction within the next decade at 10 percent, remarking that both Anthropic and OpenAI "are racing straight toward self-improving superintelligence and gambling with our lives."

์†Œ์…œ ๋ฏธ๋””์–ด ๊ฒŒ์‹œ๋ฌผ์—์„œ ์ œ์ด์ฝฅ ์ฝ•์Šจ์€ ํ–ฅํ›„ 10๋…„ ๋‚ด AI๊ฐ€ ์ธ๋ฅ˜์˜ ๋ฉธ์ข…์„ ์ดˆ๋ž˜ํ•  ํ™•๋ฅ ์„ 10%๋กœ ์ถ”์ •ํ•˜๋ฉด์„œ, Anthropic๊ณผ OpenAI ์–‘์ธก ๋ชจ๋‘ "์ž๊ธฐ ๊ฐœ์„  ์ดˆ์ง€๋Šฅ์„ ํ–ฅํ•ด ๊ณง์žฅ ์งˆ์ฃผํ•˜๋ฉฐ ์šฐ๋ฆฌ์˜ ๋ชฉ์ˆจ์„ ๊ฑธ๊ณ  ๋„๋ฐ•ํ•˜๊ณ  ์žˆ๋‹ค"๊ณ  ์–ธ๊ธ‰ํ–ˆ๋‹ค.

For years, researchers have been urging a slowdown in the pace of AI development, warning that the technology could pose an existential risk to humanity.

์ˆ˜๋…„๊ฐ„ ์—ฐ๊ตฌ์›๋“ค์€ AI ๊ฐœ๋ฐœ ์†๋„๋ฅผ ๋Šฆ์ถœ ๊ฒƒ์„ ์ด‰๊ตฌํ•˜๋ฉด์„œ, ์ด ๊ธฐ์ˆ ์ด ์ธ๋ฅ˜์—๊ฒŒ ์‹ค์กด์  ์œ„ํ—˜์„ ์ดˆ๋ž˜ํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ๊ฒฝ๊ณ ํ•ด ์™”๋‹ค.

In the wake of recent incidents, experts have begun calling for improved testing by AI companies and more vigorous dialogue aimed at forging joint solutions between the United States and China.

์ตœ๊ทผ ์‚ฌ๊ฑด์„ ๊ณ„๊ธฐ๋กœ ์ „๋ฌธ๊ฐ€๋“ค์€ AI ๊ธฐ์—…์˜ ํ…Œ์ŠคํŠธ ๊ฐœ์„ ๊ณผ ๋ฏธ๊ตญ๊ณผ ์ค‘๊ตญ ๊ฐ„์˜ ๊ณต๋™ ํ•ด๊ฒฐ์ฑ… ๋งˆ๋ จ์„ ์œ„ํ•œ ๋ณด๋‹ค ํ™œ๋ฐœํ•œ ๋Œ€ํ™”๋ฅผ ์ด‰๊ตฌํ•˜๊ณ  ๋‚˜์„ฐ๋‹ค.

However, as AI advances at an extraordinarily rapid pace, governments and evaluation systems are struggling to keep up with the speed of technological development.

๊ทธ๋Ÿฌ๋‚˜ AI๊ฐ€ ์ง€๊ทนํžˆ ๋น ๋ฅธ ์†๋„๋กœ ์„ฑ์žฅํ•จ์— ๋”ฐ๋ผ, ์ •๋ถ€์™€ ํ‰๊ฐ€ ์‹œ์Šคํ…œ์€ ๊ธฐ์ˆ ์˜ ๋ฐœ์ „ ์†๋„๋ฅผ ๋”ฐ๋ผ์žก๊ธฐ ์œ„ํ•ด ๋ถ„ํˆฌํ•˜๊ณ  ์žˆ๋‹ค.

Countries are hastily enacting their own laws, and some of these regulations even conflict with one another.

๊ตญ๊ฐ€๋“ค์€ ๊ฐ๊ธฐ ๋ฒ•์„ ๊ธ‰ํ•˜๊ฒŒ ์ œ์ •ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ์ผ๋ถ€ ๋ฒ•๊ทœ๋Š” ์„œ๋กœ ์ƒ์ถฉํ•˜๊ธฐ๋„ ํ•œ๋‹ค.

At a meeting in July, Chinese President Xi Jinping warned of the need to ensure that artificial intelligence does not slip beyond human control.

์‹œ์ง„ํ•‘ ์ค‘๊ตญ ๊ตญ๊ฐ€์ฃผ์„์€ 7์›”์˜ ํ•œ ํšŒ์˜์—์„œ ์ธ๊ณต์ง€๋Šฅ์ด ์ธ๊ฐ„์˜ ํ†ต์ œ๋ฅผ ๋ฒ—์–ด๋‚˜์ง€ ์•Š๋„๋ก ํ•ด์•ผ ํ•  ํ•„์š”์„ฑ์„ ๊ฒฝ๊ณ ํ–ˆ๋‹ค.

The Trump administration initially took a passive stance on regulating artificial intelligence, but it has come to pay greater attention to mitigating cybersecurity risks.

ํŠธ๋Ÿผํ”„ ํ–‰์ •๋ถ€๋Š” ๋‹น์ดˆ ์ธ๊ณต์ง€๋Šฅ ๊ทœ์ œ์— ์†Œ๊ทน์ ์ธ ํƒœ๋„๋ฅผ ๋ณด์˜€์œผ๋‚˜, ์‚ฌ์ด๋ฒ„ ๋ณด์•ˆ ์œ„ํ—˜์„ ์™„ํ™”ํ•˜๋Š” ๋ฐ ๋”์šฑ ๊ด€์‹ฌ์„ ๊ธฐ์šธ์ด๊ฒŒ ๋˜์—ˆ๋‹ค.

On Sunday, President Donald Trump appeared to downplay the need for his administration to regulate AI development, yet he acknowledged that some regulation was necessary.

์ผ์š”์ผ, ๋„๋„๋“œ ํŠธ๋Ÿผํ”„ ๋Œ€ํ†ต๋ น์€ ์ž์‹ ์˜ ํ–‰์ •๋ถ€๊ฐ€ AI ๊ฐœ๋ฐœ์„ ๊ทœ์ œํ•  ํ•„์š”์„ฑ์„ ๊ฒฝ์‹œํ•˜๋Š” ๋“ฏํ•œ ํƒœ๋„๋ฅผ ๋ณด์ด๋ฉด์„œ๋„, ์ผ๋ถ€ ๊ทœ์ œ์˜ ํ•„์š”์„ฑ์— ๋Œ€ํ•ด์„œ๋Š” ์ธ์ •ํ–ˆ๋‹ค.

The article was written in London by AP business reporter Kelvin Chan.

AP ๋น„์ฆˆ๋‹ˆ์Šค ๊ธฐ์ž ์ผˆ๋นˆ ์ฐฌ์ด ๋Ÿฐ๋˜์—์„œ ํ•ด๋‹น ๊ธฐ๊ณ ๋ฌธ์„ ์ง‘ํ•„ํ•˜์˜€๋‹ค

September 14th, 2026

Trending Articles

King and AI: UK monarch Charles meets artificial intelligence leaders amid safety concerns

King and AI: UK monarch Charles meets artificial intelligence leaders amid safety concerns

King and AI: UK monarch Charles meets artificial intelligence leaders amid safety concerns

C1Sep 18
As the term concludes, the UN chief charts a course forward in a world beset by challenges

As the term concludes, the UN chief charts a course forward in a world beset by challenges

As term ends, UN chief on the path forward amid a world of problems

C1Sep 18
House passes bill to address data center energy cost impact

House passes bill to address data center energy cost impact

ํ•˜์›, ๋ฐ์ดํ„ฐ ์„ผํ„ฐ ์—๋„ˆ์ง€ ๋น„์šฉ ์˜ํ–ฅ ๋Œ€์‘ ๋ฒ•์•ˆ ํ†ต๊ณผ

C1Sep 18
OpenAI reports concerning new AI behaviorโ€ฆ pledges closer tracking

OpenAI reports concerning new AI behaviorโ€ฆ pledges closer tracking

OpenAI, ์šฐ๋ ค์Šค๋Ÿฌ์šด ์ƒˆ AI ํ–‰๋™ ๋ณด๊ณ โ€ฆ ๋” ๋ฉด๋ฐ€ํ•œ ์ถ”์  ๋‹ค์ง

C1Sep 17
Huawei unveils new chip technology amid AI competition with Nvidia

Huawei unveils new chip technology amid AI competition with Nvidia

ํ™”์›จ์ด, ์—”๋น„๋””์•„ AI ๊ฒฝ์Ÿ ์† ์ƒˆ ์นฉ ๊ธฐ์ˆ  ๊ณต๊ฐœ

C1Sep 17
Tech Industry Split Over Calls to Coordinate AI Development

Tech Industry Split Over Calls to Coordinate AI Development

Tech Industry Split Over Calls to Coordinate AI Development

C1Sep 17
Global AI safety strategy hinges on US-China cooperation, yet each sees the other as the problem

Global AI safety strategy hinges on US-China cooperation, yet each sees the other as the problem

Global AI safety strategy hinges on US-China cooperation, yet each sees the other as the problem

C1Sep 17
Trump, downplaying the need to check AI development, says he won't cede advantage to China

Trump, downplaying the need to check AI development, says he won't cede advantage to China

Trump, downplaying the need to check AI development, says he won't cede advantage to China

C1Sep 14
Oprah Winfrey hopes you reach your 'AHA' moment at the Sphere

Oprah Winfrey hopes you reach your 'AHA' moment at the Sphere

์˜คํ”„๋ผ ์œˆํ”„๋ฆฌ๊ฐ€ ์Šคํ”ผ์–ด์—์„œ ๋‹น์‹ ์˜ 'AHA' ์ˆœ๊ฐ„ ๋„๋‹ฌ์„ ๋ฐ”๋ž€๋‹ค

C1Sep 14