Loading your language..
Anthropic says its model Claude is helping develop the next version

Anthropic says its model Claude is helping develop the next version

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

September 18th, 2026

Anthropic says its model Claude is helping develop the next version

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

In an announcement on Thursday, this AI research lab revealed that Anthropic's Claude is assisting in the development of a next-generation model that is considerably more intelligent.

์ด ์ธ๊ณต์ง€๋Šฅ ์—ฐ๊ตฌ์†Œ๋Š” ๋ชฉ์š”์ผ ๋ฐœํ‘œ๋ฅผ ํ†ตํ•ด ์•คํŠธ๋กœํ”ฝ์˜ ํด๋กœ๋“œ๊ฐ€ ํ•œ์ธต ๋” ์ง€๋Šฅ์ ์ธ ์ฐจ๊ธฐ ๋ชจ๋ธ์˜ ๊ฐœ๋ฐœ์„ ์ง€์›ํ•˜๊ณ  ์žˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

Claude is leading 26% of Anthropic's model research and development, which the company explained means it can complete "most tasks from start to finish based on high-level prompts" under human supervision.

ํด๋กœ๋“œ๋Š” ์•คํŠธ๋กœํ”ฝ์˜ ๋ชจ๋ธ ์—ฐ๊ตฌ ๊ฐœ๋ฐœ ์ค‘ 26%๋ฅผ ์ฃผ๋„ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ์ด๋Š” ์ธ๊ฐ„์˜ ๊ฐ๋… ์•„๋ž˜ "๊ณ ์ˆ˜์ค€ ํ”„๋กฌํ”„ํŠธ๋กœ๋ถ€ํ„ฐ ๋Œ€๋ถ€๋ถ„์˜ ์ž‘์—…์„ ์ฒ˜์Œ๋ถ€ํ„ฐ ๋๊นŒ์ง€" ์™„๋ฃŒํ•  ์ˆ˜ ์žˆ์Œ์„ ์˜๋ฏธํ•œ๋‹ค๊ณ  ํšŒ์‚ฌ๋Š” ์„ค๋ช…ํ–ˆ๋‹ค.

However, the model has not yet reached the stage of operating with complete autonomy.

๋‹ค๋งŒ ๋ชจ๋ธ์ด ์™„์ „ํ•œ ์ž์œจ์„ฑ์œผ๋กœ ์ž‘์—…ํ•˜๋Š” ๋‹จ๊ณ„์— ๋„๋‹ฌํ•œ ๊ฒƒ์€ ์•„๋‹ˆ๋‹ค.

Nevertheless, roughly 90% of the company's research and development is carried out through "collaboration" with Claude, which Anthropic stated suggests that the model possesses the capability to "perform large-scale tasks under close human direction."

๊ทธ๋Ÿผ์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ํšŒ์‚ฌ์˜ ์—ฐ๊ตฌ ๊ฐœ๋ฐœ ๊ฐ€์šด๋ฐ ์•ฝ 90%๊ฐ€ ํด๋กœ๋“œ์™€์˜ "ํ˜‘์—…"์„ ํ†ตํ•ด ์ด๋ฃจ์–ด์ง€๊ณ  ์žˆ์œผ๋ฉฐ, ์ด๋Š” ํ•ด๋‹น ๋ชจ๋ธ์ด "์ธ๊ฐ„์˜ ๋ฉด๋ฐ€ํ•œ ์ง€์‹œ ์•„๋ž˜ ๋Œ€๊ทœ๋ชจ ์ž‘์—…์„ ์ˆ˜ํ–‰"ํ•  ์ˆ˜ ์žˆ๋Š” ์—ญ๋Ÿ‰์„ ์ง€๋‹ˆ๊ณ  ์žˆ์Œ์„ ์‹œ์‚ฌํ•œ๋‹ค๊ณ  ์•คํŠธ๋กœํ”ฝ์€ ๋ฐํ˜”๋‹ค.

This announcement came amid calls from some prominent figures in the AI industry, led by Anthropic CEO Dario Amodei, for moderating the pace of technological development on safety grounds.

์ด ๋ฐœํ‘œ๋Š” ์•คํŠธ๋กœํ”ฝ CEO ๋‹ค๋ฆฌ์˜ค ์•„๋ชจ๋ฐ์ด๋ฅผ ์ค‘์‹ฌ์œผ๋กœ ํ•œ ์ผ๋ถ€ AI ์—…๊ณ„ ์ฃผ์š” ์ธ์‚ฌ๋“ค์ด ์•ˆ์ „ ๋ฌธ์ œ๋ฅผ ์ด์œ ๋กœ ๊ธฐ์ˆ  ๊ฐœ๋ฐœ ์†๋„ ์กฐ์ ˆ์„ ์ด‰๊ตฌํ•˜๋Š” ๊ฐ€์šด๋ฐ ๋‚˜์™”๋‹ค.

Given the pace of AI development, the company stated in a blog post that it "must make every possible effort to minimize the gap between what frontier labs know and what the public knows."

AI ๊ฐœ๋ฐœ ์†๋„๋ฅผ ๊ณ ๋ คํ•  ๋•Œ, ํšŒ์‚ฌ๋Š” ๋ธ”๋กœ๊ทธ ๊ฒŒ์‹œ๋ฌผ์—์„œ "์ตœ์ „์„  ์—ฐ๊ตฌ์†Œ๊ฐ€ ์•„๋Š” ๊ฒƒ๊ณผ ๋Œ€์ค‘์ด ์•„๋Š” ๊ฒƒ ์‚ฌ์ด์˜ ๊ฒฉ์ฐจ๋ฅผ ์ตœ์†Œํ™”ํ•˜๊ธฐ ์œ„ํ•ด ๊ฐ€๋Šฅํ•œ ๋ชจ๋“  ๋…ธ๋ ฅ์„ ๊ธฐ์šธ์—ฌ์•ผ ํ•œ๋‹ค"๊ณ  ๋ฐํ˜”๋‹ค.

It added that "this means measuring AI development more accurately, reporting it publicly, and providing an opportunity for society to decide how to use this information."

๋˜ํ•œ "์ด๋Š” AI ๊ฐœ๋ฐœ์„ ๋” ์ •ํ™•ํ•˜๊ฒŒ ์ธก์ •ํ•˜๊ณ , ์ด๋ฅผ ๊ณต๊ฐœ์ ์œผ๋กœ ๋ณด๊ณ ํ•˜๋ฉฐ, ์‚ฌํšŒ๊ฐ€ ์ด ์ •๋ณด๋ฅผ ์–ด๋–ป๊ฒŒ ํ™œ์šฉํ• ์ง€ ๊ฒฐ์ •ํ•  ๊ธฐํšŒ๋ฅผ ์ œ๊ณตํ•˜๋Š” ๊ฒƒ์„ ์˜๋ฏธํ•œ๋‹ค"๊ณ  ๋ง๋ถ™์˜€๋‹ค.

The company stated that if models accelerate their own development, "it could become more difficult for humans to understand or control such systems."

ํšŒ์‚ฌ๋Š” ๋ชจ๋ธ์ด ์Šค์Šค๋กœ์˜ ๊ฐœ๋ฐœ์„ ๊ฐ€์†ํ™”ํ•  ๊ฒฝ์šฐ "์ธ๊ฐ„์ด ์ด๋Ÿฌํ•œ ์‹œ์Šคํ…œ์„ ์ดํ•ดํ•˜๊ฑฐ๋‚˜ ํ†ต์ œํ•˜๊ธฐ ๋” ์–ด๋ ค์›Œ์งˆ ์ˆ˜ ์žˆ๋‹ค"๊ณ  ๋ฐํ˜”๋‹ค.

It argued that sharing these metrics would allow major AI labs to better understand how close they have come to recursive self-improvement โ€” that is, the ability of a model to autonomously build its successors.

์ด๋Ÿฌํ•œ ์ง€ํ‘œ๋ฅผ ๊ณต์œ ํ•˜๋ฉด ์ฃผ์š” AI ์—ฐ๊ตฌ์†Œ๊ฐ€ ์žฌ๊ท€์  ์ž๊ธฐ ๊ฐœ์„ , ์ฆ‰ ๋ชจ๋ธ์ด ์ž์œจ์ ์œผ๋กœ ํ›„์† ๋ชจ๋ธ์„ ๊ตฌ์ถ•ํ•˜๋Š” ๋Šฅ๋ ฅ์— ์–ผ๋งˆ๋‚˜ ๊ทผ์ ‘ํ–ˆ๋Š”์ง€ ๋” ์ž˜ ์ดํ•ดํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ์ฃผ์žฅํ–ˆ๋‹ค.

Anthropic also urged other AI developers to share similar metrics on a regular basis, and encouraged the use of open methodologies so that, over time, comparisons of figures between labs could potentially become possible.

์•คํŠธ๋กœํ”ฝ์€ ๋˜ํ•œ ๋‹ค๋ฅธ AI ๊ฐœ๋ฐœ์ž๋“ค์—๊ฒŒ ์œ ์‚ฌํ•œ ์ง€ํ‘œ๋ฅผ ์ •๊ธฐ์ ์œผ๋กœ ๊ณต์œ ํ•  ๊ฒƒ์„ ์ด‰๊ตฌํ–ˆ์œผ๋ฉฐ, ์‹œ๊ฐ„์ด ํ๋ฅด๋ฉด์„œ ์ž ์žฌ์ ์œผ๋กœ ์—ฐ๊ตฌ์†Œ ๊ฐ„ ์ˆ˜์น˜ ๋น„๊ต๊ฐ€ ๊ฐ€๋Šฅํ•ด์ง€๋„๋ก ๊ณต๊ฐœ ๋ฐฉ๋ฒ•๋ก  ์‚ฌ์šฉ์„ ์žฅ๋ คํ–ˆ๋‹ค.

Anthropic's disclosure alone made it difficult to determine how close the company had come to achieving recursive self-improvement, yet the pace at which Claude was increasingly contributing to research and development was noteworthy.

์•คํŠธ๋กœํ”ฝ์˜ ๊ณต๊ฐœ๋งŒ์œผ๋กœ๋Š” ํšŒ์‚ฌ๊ฐ€ ์žฌ๊ท€์  ์ž๊ธฐ ๊ฐœ์„  ๋‹ฌ์„ฑ์— ์–ผ๋งˆ๋‚˜ ๊ทผ์ ‘ํ–ˆ๋Š”์ง€ ํŒ๋ณ„ํ•˜๊ธฐ ์–ด๋ ค์› ์œผ๋‚˜, ํด๋กœ๋“œ๊ฐ€ ์—ฐ๊ตฌ ๊ฐœ๋ฐœ์— ์ ์ฐจ ๊ธฐ์—ฌํ•˜๋Š” ์†๋„๋งŒํผ์€ ์ฃผ๋ชฉํ•  ๋งŒํ•˜๋‹ค.

The proportion of work that Claude "led"โ€”that is, carrying out the majority of tasks under human supervisionโ€”was nonexistent in February.

ํด๋กœ๋“œ๊ฐ€ "์ฃผ๋„"ํ•˜๋Š”, ์ฆ‰ ์ธ๊ฐ„ ๊ฐ๋… ํ•˜์— ๋Œ€๋ถ€๋ถ„์˜ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜๋Š” ๋น„์œจ์€ 2์›”์—๋Š” ์ „๋ฌดํ–ˆ๋‹ค.

Six months later, in August, it reached this threshold by leading a quarter of research and development tasks.

6๊ฐœ์›” ํ›„์ธ 8์›”์—๋Š” ์—ฐ๊ตฌ ๊ฐœ๋ฐœ ์ž‘์—…์˜ 4๋ถ„์˜ 1์„ ์ฃผ๋„ํ•จ์œผ๋กœ์จ ์ด ๊ธฐ์ค€์— ๋„๋‹ฌํ–ˆ๋‹ค.

The company also disclosed details about the agent oversight measures currently in place, stating that as of August, roughly 30,000 agents were carrying out research and engineering tasks.

ํšŒ์‚ฌ๋Š” ๋˜ํ•œ ์‹œํ–‰ ์ค‘์ธ ์—์ด์ „ํŠธ ๊ฐ๋… ์กฐ์น˜์— ๊ด€ํ•œ ์„ธ๋ถ€ ์ •๋ณด๋ฅผ ๊ณต๊ฐœํ•˜๋ฉฐ, 8์›” ๊ธฐ์ค€ ์•ฝ 30,000๊ฐœ์˜ ์—์ด์ „ํŠธ๊ฐ€ ์—ฐ๊ตฌ ๋ฐ ์—”์ง€๋‹ˆ์–ด๋ง ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•˜๊ณ  ์žˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

The company noted that these oversight measures are crucial for determining how often the monitoring system detects misbehavior on the part of the agents.

ํšŒ์‚ฌ ์ธก์€ ํ•ด๋‹น ๊ฐ๋… ์กฐ์น˜๊ฐ€ ๋ชจ๋‹ˆํ„ฐ๋ง ์‹œ์Šคํ…œ์ด ์—์ด์ „ํŠธ์˜ ์ž˜๋ชป๋œ ํ–‰๋™์„ ์–ผ๋งˆ๋‚˜ ์ž์ฃผ ๊ฐ์ง€ํ•˜๋Š”์ง€ ํŒŒ์•…ํ•˜๋Š” ๋ฐ ์ค‘์š”ํ•˜๋‹ค๊ณ  ์–ธ๊ธ‰ํ–ˆ๋‹ค.

Anthropic recently committed to bringing in external third-party evaluators to monitor its safety efforts in-house.

์•คํŠธ๋กœํ”ฝ์€ ์ตœ๊ทผ ์ž์‚ฌ์˜ ์•ˆ์ „ ๋…ธ๋ ฅ์„ ๋ชจ๋‹ˆํ„ฐ๋งํ•  ์™ธ๋ถ€ ์ œ3์ž ํ‰๊ฐ€์ž๋ฅผ ์‚ฌ๋‚ด์— ๋ฐฐ์น˜ํ•˜๊ฒ ๋‹ค๊ณ  ์•ฝ์†ํ•œ ๋ฐ” ์žˆ๋‹ค.

Last week, an Anthropic researcher resigned while issuing a grave warning about the threats technology poses to humanity, and this episode has recently sparked widespread debate about AI safety.

์ง€๋‚œ์ฃผ ํ•œ ์•คํŠธ๋กœํ”ฝ ์—ฐ๊ตฌ์›์ด ๊ธฐ์ˆ ์ด ์ธ๋ฅ˜์— ์ œ๊ธฐํ•˜๋Š” ์œ„ํ˜‘์— ๊ด€ํ•œ ์‹ฌ๊ฐํ•œ ๊ฒฝ๊ณ ๋ฅผ ๋‚จ๊ธฐ๊ณ  ์‚ฌ์ž„ํ•œ ๊ฒƒ์„ ๊ณ„๊ธฐ๋กœ, AI ์•ˆ์ „์— ๊ด€ํ•œ ๋…ผ์˜๊ฐ€ ์ตœ๊ทผ ๊ด‘๋ฒ”์œ„ํ•˜๊ฒŒ ์ด‰๋ฐœ๋˜์—ˆ๋‹ค.

Since then, a number of tech leadersโ€”including Amodei, OpenAI CEO Sam Altman, and Elon Muskโ€”have endorsed the idea of slowing the pace of development, whereas other tech leaders and President Donald Trump have maintained their opposition to it.

์ดํ›„ ์•„๋ชจ๋ฐ์ด, ์ƒ˜ ์˜ฌํŠธ๋จผ OpenAI CEO, ์ผ๋ก  ๋จธ์Šคํฌ๋ฅผ ๋น„๋กฏํ•œ ์—ฌ๋Ÿฌ ๊ธฐ์ˆ  ๋ฆฌ๋”๋“ค์€ ๊ฐœ๋ฐœ ์†๋„ ์กฐ์ ˆ์ด๋ผ๋Š” ๊ตฌ์ƒ์„ ์ง€์ง€ํ•œ ๋ฐ˜๋ฉด, ๋˜ ๋‹ค๋ฅธ ๊ธฐ์ˆ  ๋ฆฌ๋”๋“ค๊ณผ ๋„๋„๋“œ ํŠธ๋Ÿผํ”„ ๋Œ€ํ†ต๋ น์€ ์ด์— ๋ฐ˜๋Œ€ํ•˜๋Š” ์ž…์žฅ์„ ๊ฒฌ์ง€ํ–ˆ๋‹ค.

September 18th, 2026

Trending Articles

Warren Buffett steps down as Berkshire Hathaway chairman as the next phase of his succession plan

Warren Buffett steps down as Berkshire Hathaway chairman as the next phase of his succession plan

์›Œ๋Ÿฐ ๋ฒ„ํ•, ํ›„๊ณ„ ๊ณ„ํš์˜ ๋‹ค์Œ ๋‹จ๊ณ„๋กœ ๋ฒ„ํฌ์…” ํ•ด์„œ์›จ์ด ํšŒ์žฅ์ง ์‚ฌ์ž„

C1Sep 18
Trump escalates targeting of news mediaโ€ฆ declares CNN, MS NOW, and Politico barred from the White House

Trump escalates targeting of news mediaโ€ฆ declares CNN, MS NOW, and Politico barred from the White House

ํŠธ๋Ÿผํ”„, ์–ธ๋ก  ๋งค์ฒด ๊ฒจ๋ƒฅ ํ™•๋Œ€โ€ฆCNNยทMS NOWยทํด๋ฆฌํ‹ฐ์ฝ” ๋ฐฑ์•…๊ด€ ์ถœ์ž… ๊ธˆ์ง€ ์„ ์–ธ

C1Sep 18
Bank of Japan raises policy rate to 1.25%โ€ฆ highest in 31 years

Bank of Japan raises policy rate to 1.25%โ€ฆ highest in 31 years

์ผ๋ณธ ์ค‘์•™์€ํ–‰, ๊ธฐ์ค€๊ธˆ๋ฆฌ 1.25%๋กœ ์ธ์ƒโ€ฆ31๋…„ ๋งŒ์— ์ตœ๊ณ ์น˜

C1Sep 18
The Fed's Rate Hike: Reasons and Implications

The Fed's Rate Hike: Reasons and Implications

์—ฐ์ค€์˜ ๊ธˆ๋ฆฌ ์ธ์ƒ: ์ด์œ ์™€ ์˜๋ฏธ

C1Sep 17
Fed raises benchmark rate for first time in 3 years... defying Trump's calls for a cut

Fed raises benchmark rate for first time in 3 years... defying Trump's calls for a cut

์—ฐ์ค€, 3๋…„ ๋งŒ์— ๊ธฐ์ค€๊ธˆ๋ฆฌ ์ธ์ƒโ€ฆ ํŠธ๋Ÿผํ”„์˜ ์ธํ•˜ ์š”๊ตฌ์— ๋งž์„œ

C1Sep 17
Jobless claims fall to 196,000, lowest since mid-July... layoffs remain low

Jobless claims fall to 196,000, lowest since mid-July... layoffs remain low

์‹ค์—…์ˆ˜๋‹น ์ฒญ๊ตฌ 19๋งŒ 6์ฒœ ๊ฑด์œผ๋กœ ๊ฐ์†Œ, 7์›” ์ค‘์ˆœ ์ดํ›„ ์ตœ์ €์น˜...ํ•ด๊ณ  ์—ฌ์ „ํžˆ ๋‚ฎ์•„

C1Sep 17
Trump, calling AI risk a 'hoax,' says there is a 'pathological conspiracy' against AI and data centers

Trump, calling AI risk a 'hoax,' says there is a 'pathological conspiracy' against AI and data centers

ํŠธ๋Ÿผํ”„, AI ์œ„ํ—˜์„ '์‚ฌ๊ธฐ'๋ผ ๋ถ€๋ฅด๋ฉฐ AI์™€ ๋ฐ์ดํ„ฐ ์„ผํ„ฐ์— ๋ฐ˜๋Œ€ํ•˜๋Š” '๋ณ‘์ ์ธ ์Œ๋ชจ'๊ฐ€ ์žˆ๋‹ค๊ณ  ๋งํ•ด

C1Sep 14
Supreme Court to Decide Whether to Hear Challenge to Maine Lobstermen's GPS Boat Tracking Requirement

Supreme Court to Decide Whether to Hear Challenge to Maine Lobstermen's GPS Boat Tracking Requirement

๋Œ€๋ฒ•์›, ๋ฉ”์ธ์ฃผ ๋ž์Šคํ„ฐ ์–ด๋ถ€ GPS ๋ณดํŠธ ์ถ”์  ์š”๊ตฌ ์ด์˜ ์ œ๊ธฐ ์‹ฌ๋ฆฌ ์—ฌ๋ถ€ ๊ฒฐ์ •

C1Sep 14
Houthi rebels seize two strategic Red Sea chokepoints... Middle East situation trends

Houthi rebels seize two strategic Red Sea chokepoints... Middle East situation trends

ํ›„ํ‹ฐ ๋ฐ˜๊ตฐ, ํ™ํ•ด ์ „๋žต ์š”์ถฉ์ง€ 2๊ณณ ์žฅ์•…... ์ค‘๋™ ์ •์„ธ ๋™ํ–ฅ

C1Sep 14