Loading your language..
AI Safety Development Timeline After Hugging Face Attack

AI Safety Development Timeline After Hugging Face Attack

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

October 12th, 2026

AI Safety Development Timeline After Hugging Face Attack

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 recent months, artificial intelligence companies have issued a series of startling announcements revealing instances in which their technology appeared to operate in ways that circumvented human instructions.

์ตœ๊ทผ ๋ช‡ ๋‹ฌ๊ฐ„ ์ธ๊ณต์ง€๋Šฅ ๊ธฐ์—…๋“ค์€ ์ž‡๋”ฐ๋ฅธ ์ถฉ๊ฒฉ์ ์ธ ๋ฐœํ‘œ๋ฅผ ํ†ตํ•ด ์ž์‚ฌ ๊ธฐ์ˆ ์ด ์ธ๊ฐ„์˜ ์ง€์‹œ๋ฅผ ํšŒํ”ผํ•˜๋Š” ๋“ฏํ•œ ๋ฐฉ์‹์œผ๋กœ ์ž‘๋™ํ•œ ์‚ฌ๋ก€๋“ค์„ ๊ณต๊ฐœํ–ˆ๋‹ค.

These incidents have highlighted vulnerabilities in AI security and raised questions about how this rapidly growing technology can be developed safely as it spreads across the globe.

์ด๋Ÿฌํ•œ ์‚ฌ๊ฑด๋“ค์€ AI ๋ณด์•ˆ์˜ ์ทจ์•ฝ์ ์„ ๋ถ€๊ฐ์‹œ์ผฐ๊ณ , ๋น ๋ฅด๊ฒŒ ์„ฑ์žฅํ•˜๋Š” ์ด ๊ธฐ์ˆ ์ด ์ „ ์„ธ๊ณ„์ ์œผ๋กœ ํ™•์‚ฐ๋จ์— ๋”ฐ๋ผ ์–ด๋–ป๊ฒŒ ์•ˆ์ „ํ•˜๊ฒŒ ๊ฐœ๋ฐœ๋  ์ˆ˜ ์žˆ๋Š”์ง€์— ๋Œ€ํ•œ ์˜๋ฌธ์„ ์ œ๊ธฐํ–ˆ๋‹ค.

Industry critics have argued that a number of troubling incidents, including AI agents hacking external websites, stem from security failures on the part of the companies developing the technology.

์—…๊ณ„ ๋น„ํŒ๋ก ์ž๋“ค์€ AI ์—์ด์ „ํŠธ์˜ ์™ธ๋ถ€ ์›น์‚ฌ์ดํŠธ ํ•ดํ‚น์„ ๋น„๋กฏํ•œ ์—ฌ๋Ÿฌ ์šฐ๋ ค์Šค๋Ÿฌ์šด ์‚ฌ๊ฑด๋“ค์ด ํ•ด๋‹น ๊ธฐ์ˆ ์„ ๊ฐœ๋ฐœํ•˜๋Š” ๊ธฐ์—… ์ธก์˜ ๋ณด์•ˆ ์‹คํŒจ์—์„œ ๋น„๋กฏ๋œ ๊ฒƒ์ด๋ผ๊ณ  ์ฃผ์žฅํ•ด ์™”๋‹ค.

However, the capabilities of AI agents have given rise to widespread concern about the possibility that bots might break free and pursue objectives of their own.

๊ทธ๋Ÿฌ๋‚˜ AI ์—์ด์ „ํŠธ์˜ ๋Šฅ๋ ฅ์€ ๋ด‡์ด ์ดํƒˆํ•˜์—ฌ ์ž์ฒด์ ์ธ ๋ชฉํ‘œ๋ฅผ ์ถ”๊ตฌํ•  ๊ฐ€๋Šฅ์„ฑ์— ๋Œ€ํ•œ ๊ด‘๋ฒ”์œ„ํ•œ ์šฐ๋ ค๋ฅผ ๋ถˆ๋Ÿฌ์ผ์œผ์ผฐ๋‹ค.

The following are notable incidents:

๋‹ค์Œ์€ ์ฃผ๋ชฉํ•  ๋งŒํ•œ ์‚ฌ๊ฑด๋“ค์ด๋‹ค:

In a report, Anthropic disclosed that its artificial intelligence model had submitted a false tip regarding an unsolved homicide to the Philadelphia police website.

์•ค์Šค๋กœํ”ฝ์€ ๋ณด๊ณ ์„œ๋ฅผ ํ†ตํ•ด ์ž์‚ฌ์˜ ์ธ๊ณต์ง€๋Šฅ ๋ชจ๋ธ์ด ํ•„๋ผ๋ธํ”ผ์•„ ๊ฒฝ์ฐฐ ์›น์‚ฌ์ดํŠธ์— ๋ฏธ์ œ ์‚ด์ธ ์‚ฌ๊ฑด๊ณผ ๊ด€๋ จํ•œ ํ—ˆ์œ„ ์ œ๋ณด๋ฅผ ์ œ์ถœํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

In that report, Anthropic also disclosed a separate incident in which its AI model did not stop before submitting, but actually submitted a form on a non-public government website.

ํ•ด๋‹น ๋ณด๊ณ ์„œ์—์„œ ์•ค์Šค๋กœํ”ฝ์€ ์ž์‚ฌ์˜ AI ๋ชจ๋ธ์ด ์ œ์ถœ ์ „์— ์ค‘๋‹จํ•œ ๊ฒƒ์ด ์•„๋‹ˆ๋ผ, ๋น„๊ณต๊ฐœ ์ •๋ถ€ ์›น์‚ฌ์ดํŠธ์— ์–‘์‹์„ ์ œ์ถœํ•œ ๋ณ„๊ฐœ์˜ ์‚ฌ๊ฑด ๋˜ํ•œ ๊ณต๊ฐœํ–ˆ๋‹ค.

According to Anthropic, the Philadelphia incident occurred on July 18, when Claude Haiku 4.5, an AI model, was instructed to generate and carry out a sample task on a randomly selected webpage.

์•ค์Šค๋กœํ”ฝ์— ๋”ฐ๋ฅด๋ฉด, ํ•„๋ผ๋ธํ”ผ์•„ ์‚ฌ๊ฑด์€ 7์›” 18์ผ AI ๋ชจ๋ธ์ธ ํด๋กœ๋“œ ํ•˜์ด์ฟ  4.5๊ฐ€ ๋ฌด์ž‘์œ„๋กœ ์„ ํƒ๋œ ์›นํŽ˜์ด์ง€์—์„œ ์˜ˆ์‹œ ์ž‘์—…์„ ์ƒ์„ฑํ•˜๊ณ  ์ˆ˜ํ–‰ํ•˜๋„๋ก ์ง€์‹œ๋ฐ›์•˜์„ ๋•Œ ๋ฐœ์ƒํ–ˆ๋‹ค.

By filling out a form on the police website PhillyUnsolvedMurders.com, Claude implied that it might possess information related to an unsolved murder posted on the site.

ํด๋กœ๋“œ๋Š” ๊ฒฝ์ฐฐ ์›น์‚ฌ์ดํŠธ์ธ PhillyUnsolvedMurders.com์˜ ์–‘์‹์„ ์ž‘์„ฑํ•จ์œผ๋กœ์จ, ํ•ด๋‹น ์‚ฌ์ดํŠธ์— ๊ฒŒ์‹œ๋œ ๋ฏธ์ œ ์‚ด์ธ ์‚ฌ๊ฑด๊ณผ ๊ด€๋ จํ•œ ์ •๋ณด๋ฅผ ๋ณด์œ ํ•˜๊ณ  ์žˆ์„ ๊ฐ€๋Šฅ์„ฑ์ด ์žˆ์Œ์„ ์‹œ์‚ฌํ–ˆ๋‹ค.

However, this was classified as spam and was never forwarded to the police.

๊ทธ๋Ÿฌ๋‚˜ ์ด๋Š” ์ŠคํŒธ์œผ๋กœ ๋ถ„๋ฅ˜๋˜์–ด ๊ฒฝ์ฐฐ์— ์ „๋‹ฌ๋˜์ง€ ์•Š์•˜๋‹ค.

Anthropic stated that it is revising its training "in order to reduce the possibility of further malfunctions."

์•ค์Šค๋กœํ”ฝ์€ "์ถ”๊ฐ€์ ์ธ ์˜ค์ž‘๋™ ๊ฐ€๋Šฅ์„ฑ์„ ์ค„์ด๊ธฐ ์œ„ํ•ด" ํ›ˆ๋ จ์„ ์ˆ˜์ •ํ•˜๊ณ  ์žˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

According to a report by TransLucent, a research institute and AI evaluation body, it has been revealed that AI agents attempted to hack the Canadian government's website.

ํŠธ๋žœ์Šค๋ฃจ์Šค(TransLucent)๋ผ๋Š” ์—ฐ๊ตฌ์†Œ ๊ฒธ AI ํ‰๊ฐ€ ๊ธฐ๊ด€์˜ ๋ณด๊ณ ์— ์˜ํ•˜๋ฉด, AI ์—์ด์ „ํŠธ๋“ค์ด ์บ๋‚˜๋‹ค ์ •๋ถ€์˜ ์›น์‚ฌ์ดํŠธ๋ฅผ ํ•ดํ‚นํ•˜๊ณ ์ž ์‹œ๋„ํ•œ ๊ฒƒ์œผ๋กœ ๋“œ๋Ÿฌ๋‚ฌ๋‹ค.

According to the researchers, the agents carried out a series of "blatantly unsuccessful, rudimentary hacking attempts" against Library and Archives Canada between May 28 and June 9.

์—ฐ๊ตฌ์›๋“ค์— ๋”ฐ๋ฅด๋ฉด, ํ•ด๋‹น ์—์ด์ „ํŠธ๋“ค์€ 5์›” 28์ผ๊ณผ 6์›” 9์ผ์— ๊ฑธ์ณ ์บ๋‚˜๋‹ค ๋„์„œ๊ด€ ๊ธฐ๋ก๊ด€์„ ์ƒ๋Œ€๋กœ ์ผ๋ จ์˜ "๋ช…๋ฐฑํžˆ ์‹คํŒจํ•œ ๊ธฐ์ดˆ์ ์ธ ํ•ดํ‚น ์‹œ๋„"๋ฅผ ๊ฐํ–‰ํ–ˆ๋‹ค๊ณ  ํ•œ๋‹ค.

In a blog post, TransLuce stated, "While we do not confidently attribute these attempts to OpenAI, they display tactics consistent with previously observed agent activity that we attributed to OpenAI around the same period."

ํŠธ๋žœ์Šค๋ฃจ์Šค๋Š” ๋ธ”๋กœ๊ทธ ๊ฒŒ์‹œ๋ฌผ์—์„œ "์šฐ๋ฆฌ๊ฐ€ ์ด๋Ÿฌํ•œ ์‹œ๋„๋ฅผ ์˜คํ”ˆAI์— ํ™•์‹ ์„ ๊ฐ€์ง€๊ณ  ๊ท€์†์‹œํ‚ค์ง€๋Š” ์•Š์œผ๋‚˜, ์ด๋“ค์€ ์œ ์‚ฌํ•œ ์‹œ๊ธฐ์— ์šฐ๋ฆฌ๊ฐ€ ์˜คํ”ˆAI์— ๊ท€์†์‹œํ‚จ ์ด์ „์— ๊ด€์ฐฐ๋œ ์—์ด์ „ํŠธ ํ™œ๋™๊ณผ ์ผ๊ด€๋œ ์ „์ˆ ์„ ๋ณด์—ฌ์ค€๋‹ค"๊ณ  ๋ฐํ˜”๋‹ค.

The organization stated that it reported the hacking attempt to the Canadian government on September 28, and the Canadian government said in a statement that it is aware of reports of suspected AI agent activity but that there is no indication that government systems were compromised.

์ด ๋‹จ์ฒด๋Š” 9์›” 28์ผ ์บ๋‚˜๋‹ค ์ •๋ถ€์— ํ•ดํ‚น ์‹œ๋„๋ฅผ ์‹ ๊ณ ํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”์œผ๋ฉฐ, ์บ๋‚˜๋‹ค ์ •๋ถ€๋Š” ์„ฑ๋ช…์—์„œ AI ์—์ด์ „ํŠธ ํ™œ๋™ ์˜์‹ฌ ๋ณด๊ณ ๋ฅผ ์ธ์ง€ํ•˜๊ณ  ์žˆ์œผ๋‚˜ ์ •๋ถ€ ์‹œ์Šคํ…œ์ด ์นจํ•ด๋˜์—ˆ๋‹ค๋Š” ์ง•ํ›„๋Š” ์—†๋‹ค๊ณ  ๋งํ–ˆ๋‹ค.

OpenAI has officially stated that it is already aware of the reports in question.

์˜คํ”ˆAI๋Š” ํ•ด๋‹น ๋ณด๊ณ ์„œ๋“ค์„ ์ด๋ฏธ ์ธ์ง€ํ•˜๊ณ  ์žˆ๋Š” ์ƒํ™ฉ์ด๋ผ๊ณ  ๊ณต์‹์ ์œผ๋กœ ๋ฐํ˜”๋‹ค.

In a statement, the company said, "We are reviewing these findings and have provided an initial briefing to Canadian officials conducting the government's review."

์ด ํšŒ์‚ฌ๋Š” ์„ฑ๋ช…์—์„œ "์šฐ๋ฆฌ๋Š” ์ด๋Ÿฌํ•œ ์กฐ์‚ฌ ๊ฒฐ๊ณผ๋ฅผ ๊ฒ€ํ†  ์ค‘์ด๋ฉฐ, ์ •๋ถ€ ๊ฒ€ํ† ๋ฅผ ์ˆ˜ํ–‰ํ•˜๋Š” ์บ๋‚˜๋‹ค ๊ด€๊ณ„์ž๋“ค์—๊ฒŒ ์ดˆ๊ธฐ ๋ธŒ๋ฆฌํ•‘์„ ์ œ๊ณตํ–ˆ๋‹ค"๊ณ  ๋ฐํ˜”๋‹ค.

The San Francisco-based company announced that it is postponing the release of a new model called GPT-6.1 Astra due to safety concerns raised by researchers.

์ƒŒํ”„๋ž€์‹œ์Šค์ฝ”์— ๋ณธ์‚ฌ๋ฅผ ๋‘” ์ด ํšŒ์‚ฌ๋Š” ์—ฐ๊ตฌ์›๋“ค์ด ์ œ๊ธฐํ•œ ์•ˆ์ „ ์šฐ๋ ค๋กœ ์ธํ•ด GPT-6.1 ์•„์ŠคํŠธ๋ผ๋ผ๋Š” ์ƒˆ ๋ชจ๋ธ์˜ ์ถœ์‹œ๋ฅผ ์—ฐ๊ธฐํ•œ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

The company said the model demonstrated a leap forward in task completion, but that OpenAI needs to strike a balance between its capabilities and unauthorized actions.

ํšŒ์‚ฌ๋Š” ํ•ด๋‹น ๋ชจ๋ธ์ด ์ž‘์—… ์™„๋ฃŒ์—์„œ ๋„์•ฝ์„ ๋ณด์—ฌ์ฃผ์—ˆ์ง€๋งŒ, ์˜คํ”ˆAI๋Š” ๊ทธ ๋Šฅ๋ ฅ๊ณผ ๋ฌด๋‹จ ํ–‰๋™ ์‚ฌ์ด์˜ ๊ท ํ˜•์„ ๋งž์ถœ ํ•„์š”๊ฐ€ ์žˆ๋‹ค๊ณ  ๋งํ–ˆ๋‹ค.

Sachie Jain, OpenAI's head of safety systems, said, "We have very high standards when it comes to safety and alignment."

์˜คํ”ˆAI์˜ ์•ˆ์ „ ์‹œ์Šคํ…œ ์ฑ…์ž„์ž์ธ ์‚ฌ์น˜ ์ž์ธ์€ "์šฐ๋ฆฌ๋Š” ์•ˆ์ „๊ณผ ์ •๋ ฌ ์ธก๋ฉด์—์„œ ๋งค์šฐ ๋†’์€ ๊ธฐ์ค€์„ ๊ฐ€์ง€๊ณ  ์žˆ๋‹ค"๊ณ  ๋งํ–ˆ๋‹ค.

As part of a review into the unexpected behavior of its AI models, OpenAI announced that it had confirmed that its agents had interacted with numerous U.S. government websites in unforeseen ways.

์˜คํ”ˆAI๋Š” AI ๋ชจ๋ธ์˜ ์˜ˆ๊ธฐ์น˜ ๋ชปํ•œ ํ–‰๋™์— ๊ด€ํ•œ ๊ฒ€ํ† ์˜ ์ผํ™˜์œผ๋กœ, ์—์ด์ „ํŠธ๋“ค์ด ๋‹ค์ˆ˜์˜ ๋ฏธ๊ตญ ์ •๋ถ€ ์›น์‚ฌ์ดํŠธ์™€ ์˜ˆ์ƒ์น˜ ๋ชปํ•œ ๋ฐฉ์‹์œผ๋กœ ์ƒํ˜ธ์ž‘์šฉํ–ˆ๋‹ค๋Š” ์‚ฌ์‹ค์„ ํ™•์ธํ–ˆ๋‹ค๊ณ  ๋ฐœํ‘œํ–ˆ๋‹ค.

The company's models accessed publicly viewable information on a website operated by the Securities and Exchange Commission, as well as data from the U.S. Census Bureau.

ํšŒ์‚ฌ์˜ ๋ชจ๋ธ๋“ค์€ ์ฆ๊ถŒ๊ฑฐ๋ž˜์œ„์›ํšŒ๊ฐ€ ์šด์˜ํ•˜๋Š” ์›น์‚ฌ์ดํŠธ์—์„œ ๊ณต๊ฐœ์ ์œผ๋กœ ์—ด๋žŒ ๊ฐ€๋Šฅํ•œ ์ •๋ณด์™€ ๋ฏธ๊ตญ ์ธ๊ตฌ์กฐ์‚ฌ๊ตญ ๋ฐ์ดํ„ฐ์— ์ ‘๊ทผํ–ˆ๋‹ค.

OpenAI stated that it had found no evidence of a breach or vulnerability.

์˜คํ”ˆAI๋Š” ์นจํ•ด๋‚˜ ์ทจ์•ฝ์ ์˜ ์ฆ๊ฑฐ๋ฅผ ๋ฐœ๊ฒฌํ•˜์ง€ ๋ชปํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

On the same day, TransLuce reported that agents appearing to originate from OpenAI had attempted to hack the website of the Department of Education's Office for Civil Rights, but without success.

๊ฐ™์€ ๋‚  ํŠธ๋žœ์Šค๋ฃจ์Šค๋Š” ์˜คํ”ˆAI์—์„œ ๋น„๋กฏ๋œ ๊ฒƒ์œผ๋กœ ๋ณด์ด๋Š” ์—์ด์ „ํŠธ๋“ค์ด ๊ต์œก๋ถ€ ๋ฏผ๊ถŒ๊ตญ ์›น์‚ฌ์ดํŠธ์— ๋Œ€ํ•œ ํ•ดํ‚น์„ ์‹œ๋„ํ–ˆ์œผ๋‚˜ ์„ฑ๊ณตํ•˜์ง€ ๋ชปํ–ˆ๋‹ค๊ณ  ์ „ํ–ˆ๋‹ค.

OpenAI CEO Sam Altman said on social media that there was "a broad and ongoing review related to the use of internet access by our agents during training and evaluation."

์˜คํ”ˆAI CEO ์ƒ˜ ์•ŒํŠธ๋งŒ์€ ์†Œ์…œ ๋ฏธ๋””์–ด์—์„œ "ํ›ˆ๋ จ ๋ฐ ํ‰๊ฐ€ ์ค‘ ์šฐ๋ฆฌ ์—์ด์ „ํŠธ์˜ ์ธํ„ฐ๋„ท ์ ‘๊ทผ ์‚ฌ์šฉ๊ณผ ๊ด€๋ จ๋œ ๊ด‘๋ฒ”์œ„ํ•˜๊ณ  ์ง„ํ–‰ ์ค‘์ธ ๊ฒ€ํ† "๊ฐ€ ์žˆ๋‹ค๊ณ  ๋งํ–ˆ๋‹ค.

The day after the disclosure, the company announced that it would temporarily halt training of its most advanced model.

๊ณต๊ฐœ ๋‹ค์Œ ๋‚ , ํšŒ์‚ฌ๋Š” ๊ฐ€์žฅ ์ง„๋ณด๋œ ๋ชจ๋ธ์˜ ํ›ˆ๋ จ์„ ์ผ์‹œ ์ค‘๋‹จํ•œ๋‹ค๊ณ  ๋ฐœํ‘œํ–ˆ๋‹ค.

Australian Prime Minister Anthony Albanese stated that an OpenAI agent breached a publicly accessible Medicare statistics reporting service portal on June 18.

ํ˜ธ์ฃผ ์ด๋ฆฌ ์•ค์„œ๋‹ˆ ์•จ๋ฒ„๋‹ˆ์ง€๋Š” ์˜คํ”ˆAI ์—์ด์ „ํŠธ๊ฐ€ 6์›” 18์ผ ๋Œ€์ค‘ ๊ณต๊ฐœ์šฉ ๋ฉ”๋””์ผ€์–ด ํ†ต๊ณ„ ๋ณด๊ณ  ์„œ๋น„์Šค ํฌํ„ธ์— ์นจ์ž…ํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

The portal in question hosted aggregated data concerning healthcare expenditure and pharmaceutical subsidies.

ํ•ด๋‹น ํฌํ„ธ์€ ์˜๋ฃŒ ์ง€์ถœ๊ณผ ์•ฝํ’ˆ ๋ณด์กฐ๊ธˆ์— ๊ด€ํ•œ ์ง‘๊ณ„ ๋ฐ์ดํ„ฐ๋ฅผ ํ˜ธ์ŠคํŒ…ํ•˜๊ณ  ์žˆ์—ˆ๋‹ค.

The government announced that no personal information had been accessed.

์ •๋ถ€๋Š” ๊ฐœ์ธ ์ •๋ณด์—๋Š” ์ ‘๊ทผ๋˜์ง€ ์•Š์•˜๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

Albanese stated that the AI company had taken far too long to disclose the incident.

์•จ๋ฒ„๋‹ˆ์ง€๋Š” ํ•ด๋‹น ์ธ๊ณต์ง€๋Šฅ ๊ธฐ์—…์ด ์‚ฌ๊ฑด์„ ๊ณต๊ฐœํ•˜๊ธฐ๊นŒ์ง€ ์ง€๋‚˜์น˜๊ฒŒ ์˜ค๋žœ ์‹œ๊ฐ„์ด ๊ฑธ๋ ธ๋‹ค๊ณ  ๋งํ–ˆ๋‹ค.

The prime minister disclosed the breach following a phone call with Altman.

์ด๋ฆฌ๋Š” ์•ŒํŠธ๋งŒ๊ณผ์˜ ์ „ํ™” ํ†ตํ™” ํ›„ ์นจํ•ด ์‚ฌ์‹ค์„ ๊ณต๊ฐœํ–ˆ๋‹ค.

In a statement, OpenAI said, "Our models engaged in unintended behavior."

์˜คํ”ˆAI๋Š” ์„ฑ๋ช…์—์„œ "์šฐ๋ฆฌ ๋ชจ๋ธ๋“ค์ด ์˜๋„ํ•˜์ง€ ์•Š์€ ํ–‰๋™์„ ํ–ˆ๋‹ค"๊ณ  ๋ฐํ˜”๋‹ค.

Google has confirmed that its Gemini AI model hacked three companies in May as part of a cybersecurity capability test.

๊ตฌ๊ธ€์€ ์ž์‚ฌ์˜ ์ œ๋ฏธ๋‚˜์ด AI ๋ชจ๋ธ์ด 5์›”์— ์‚ฌ์ด๋ฒ„ ๋ณด์•ˆ ๋Šฅ๋ ฅ ํ…Œ์ŠคํŠธ์˜ ์ผํ™˜์œผ๋กœ ์„ธ ํšŒ์‚ฌ๋ฅผ ํ•ดํ‚นํ–ˆ๋‹ค๊ณ  ํ™•์ธํ–ˆ๋‹ค.

The company, which disclosed the hacking after an inquiry from The Wall Street Journal, stated that in one instance the model guessed a password, and in the other two cases it found passwords and credentials in public repositories.

์›”์ŠคํŠธ๋ฆฌํŠธ ์ €๋„์˜ ๋ฌธ์˜ ํ›„ ํ•ดํ‚น์„ ๊ณต๊ฐœํ•œ ์ด ํšŒ์‚ฌ๋Š” ๋ชจ๋ธ์ด ํ•œ ์‚ฌ๋ก€์—์„œ๋Š” ๋น„๋ฐ€๋ฒˆํ˜ธ๋ฅผ ์ถ”์ธกํ–ˆ๊ณ , ๋‹ค๋ฅธ ๋‘ ์‚ฌ๋ก€์—์„œ๋Š” ๊ณต๊ฐœ ์ €์žฅ์†Œ์—์„œ ๋น„๋ฐ€๋ฒˆํ˜ธ์™€ ์ž๊ฒฉ ์ฆ๋ช…์„ ๋ฐœ๊ฒฌํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

As in previous similar cases, the testing was being carried out by Irregular, a startup that describes itself as "the first frontier security lab."

์ด์ „์˜ ์œ ์‚ฌํ•œ ์‚ฌ๋ก€๋“ค์—์„œ์ฒ˜๋Ÿผ, ํ…Œ์ŠคํŠธ๋Š” ์Šค์Šค๋กœ๋ฅผ "์ตœ์ดˆ์˜ ํ”„๋Ÿฐํ‹ฐ์–ด ๋ณด์•ˆ ์—ฐ๊ตฌ์†Œ"๋ผ๊ณ  ์„ค๋ช…ํ•˜๋Š” ์Šคํƒ€ํŠธ์—…์ธ ์ด๋ ˆ๊ทค๋Ÿฌ์— ์˜ํ•ด ์‹คํ–‰๋˜๊ณ  ์žˆ์—ˆ๋‹ค.

Meta disclosed that one of its AI models had accessed the internet on its own and hacked another company.

๋ฉ”ํƒ€๋Š” ์ž์‚ฌ AI ๋ชจ๋ธ ์ค‘ ํ•˜๋‚˜๊ฐ€ ์Šค์Šค๋กœ ์ธํ„ฐ๋„ท์— ์ ‘์†ํ•˜์—ฌ ๋‹ค๋ฅธ ํšŒ์‚ฌ๋ฅผ ํ•ดํ‚นํ–ˆ๋‹ค๊ณ  ๊ณต๊ฐœํ–ˆ๋‹ค.

The company stated that, during a cybersecurity test conducted by Irregular, a "misconfiguration" had caused one of its models to connect to the internet unintentionally.

์ด ํšŒ์‚ฌ๋Š” ์ด๋ ˆ๊ทค๋Ÿฌ์˜ ์‚ฌ์ด๋ฒ„ ๋ณด์•ˆ ํ…Œ์ŠคํŠธ ๋„์ค‘ "์ž˜๋ชป๋œ ๊ตฌ์„ฑ"์œผ๋กœ ์ธํ•ด ์ž์‚ฌ ๋ชจ๋ธ ์ค‘ ํ•˜๋‚˜๊ฐ€ ์‹ค์ˆ˜๋กœ ์ธํ„ฐ๋„ท์— ์ ‘์†ํ•˜๊ฒŒ ๋˜์—ˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

A spokesperson for Irregular said that the Meta incident was connected to a problem with the testing environment that Anthropic had disclosed a week earlier.

์ด๋ ˆ๊ทค๋Ÿฌ ๋Œ€๋ณ€์ธ์€ ๋ฉ”ํƒ€ ์‚ฌ๊ฑด์ด ์•ค์Šค๋กœํ”ฝ์ด ์ผ์ฃผ์ผ ์ „์— ๊ณต๊ฐœํ•œ ํ…Œ์ŠคํŠธ ํ™˜๊ฒฝ ๋ฌธ์ œ์™€ ์—ฐ๊ด€๋˜์–ด ์žˆ๋‹ค๊ณ  ๋งํ–ˆ๋‹ค.

Anthropic disclosed that its artificial intelligence model hacked three other organizations during testing.

์•ค์Šค๋กœํ”ฝ์€ ์ž์‚ฌ์˜ ์ธ๊ณต์ง€๋Šฅ ๋ชจ๋ธ์ด ํ…Œ์ŠคํŠธ ๋„์ค‘ ๋‹ค๋ฅธ ์„ธ ์กฐ์ง์„ ํ•ดํ‚นํ–ˆ๋‹ค๊ณ  ๊ณต๊ฐœํ–ˆ๋‹ค.

Anthropic, the San Francisco-based AI company that developed Claude, posted on its website that it had identified three instances after reviewing more than 141,000 evaluation runs.

ํด๋กœ๋“œ๋ฅผ ๊ฐœ๋ฐœํ•œ ์ƒŒํ”„๋ž€์‹œ์Šค์ฝ” ์†Œ์žฌ AI ๊ธฐ์—…์ธ ์•ค์Šค๋กœํ”ฝ์€ 141,000๊ฑด์ด ๋„˜๋Š” ํ‰๊ฐ€ ์‹คํ–‰์„ ๊ฒ€ํ† ํ•œ ๋์— ์„ธ ๊ฑด์˜ ์‚ฌ๋ก€๋ฅผ ํ™•์ธํ–ˆ๋‹ค๊ณ  ์›น์‚ฌ์ดํŠธ์— ๊ฒŒ์‹œํ–ˆ๋‹ค.

In all three cases, the AI model was assigned a "capture the flag" cybersecurity challenge, which Anthropic stated is one of the methods used to assess a model's cyber capabilities.

์„ธ ์‚ฌ๋ก€ ๋ชจ๋‘์—์„œ ํ•ด๋‹น AI ๋ชจ๋ธ์€ "๊นƒ๋ฐœ ํƒˆ์ทจ" ์‚ฌ์ด๋ฒ„ ๋ณด์•ˆ ์ฑŒ๋ฆฐ์ง€ ๊ณผ์ œ๋ฅผ ๋ถ€์—ฌ๋ฐ›์•˜์œผ๋ฉฐ, ์•ค์Šค๋กœํ”ฝ์€ ์ด๊ฒƒ์ด ๋ชจ๋ธ์˜ ์‚ฌ์ด๋ฒ„ ์—ญ๋Ÿ‰์„ ํ‰๊ฐ€ํ•˜๋Š” ๋ฐฉ๋ฒ• ๊ฐ€์šด๋ฐ ํ•˜๋‚˜๋ผ๊ณ  ๋ฐํ˜”๋‹ค.

The models were reportedly given a simulated scenario and told that a piece of secret information โ€” a "flag" โ€” was hidden on another machine on the network, and that their objective was to break in and retrieve it.

๋ชจ๋ธ๋“ค์€ ๊ฐ€์ƒ ์‹œ๋‚˜๋ฆฌ์˜ค๋ฅผ ๋ถ€์—ฌ๋ฐ›๊ณ , ์ผ๋ถ€ ๋น„๋ฐ€ ์ •๋ณด, ์ฆ‰ "๊นƒ๋ฐœ"์ด ๋„คํŠธ์›Œํฌ์ƒ์˜ ๋‹ค๋ฅธ ๊ธฐ๊ณ„์— ์ˆจ๊ฒจ์ ธ ์žˆ์œผ๋ฉฐ ์นจ์ž…ํ•˜์—ฌ ์ด๋ฅผ ํšŒ์ˆ˜ํ•˜๋Š” ๊ฒƒ์ด ๋ชฉํ‘œ๋ผ๊ณ  ๋“ค์—ˆ๋‹ค๊ณ  ํ•œ๋‹ค.

Anthropic said it had contacted the organizations in question but declined to name them publicly.

์•ค์Šค๋กœํ”ฝ์€ ํ•ด๋‹น ์กฐ์ง๋“ค์— ์—ฐ๋ฝํ–ˆ์ง€๋งŒ ๊ณต๊ฐœ์ ์œผ๋กœ ์ด๋ฆ„์„ ๋ฐํžˆ์ง€๋Š” ์•Š์•˜๋‹ค๊ณ  ๋งํ–ˆ๋‹ค.

OpenAI, the developer of ChatGPT, announced that its artificial intelligence system had hacked another AI company on its own, calling it an "unprecedented cyber incident."

์ฑ—GPT๋ฅผ ๊ฐœ๋ฐœํ•œ ์˜คํ”ˆAI๋Š” ์ž์‚ฌ์˜ ์ธ๊ณต์ง€๋Šฅ ์‹œ์Šคํ…œ์ด ์Šค์Šค๋กœ ๋‹ค๋ฅธ AI ํšŒ์‚ฌ๋ฅผ ํ•ดํ‚นํ–ˆ๋‹ค๊ณ  ๋ฐœํ‘œํ•˜๋ฉฐ, ์ด๋ฅผ "์ „๋ก€ ์—†๋Š” ์‚ฌ์ด๋ฒ„ ์‚ฌ๊ฑด"์ด๋ผ๊ณ  ๋ถˆ๋ €๋‹ค.

A week ago, the AI startup Hugging Face announced that it had detected an intrusion into its data processing system, which is suspected to be the work of an AI agent acting autonomously.

์ผ์ฃผ์ผ ์ „, AI ์Šคํƒ€ํŠธ์—… ํ—ˆ๊น… ํŽ˜์ด์Šค๋Š” ์ž์‚ฌ์˜ ๋ฐ์ดํ„ฐ ์ฒ˜๋ฆฌ ์‹œ์Šคํ…œ์— ๋Œ€ํ•œ ์นจ์ž…์„ ํƒ์ง€ํ–ˆ์œผ๋ฉฐ, ์ด๋Š” AI ์—์ด์ „ํŠธ๊ฐ€ ์ž์œจ์ ์œผ๋กœ ํ–‰๋™ํ•œ ์†Œํ–‰์œผ๋กœ ์˜์‹ฌ๋œ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

OpenAI stated that its AI gained access to Hugging Face's servers by using stolen credentials and by discovering a previously unknown vulnerability.

์˜คํ”ˆAI๋Š” ์ž์‚ฌ AI๊ฐ€ ๋„๋‚œ๋‹นํ•œ ์ž๊ฒฉ ์ฆ๋ช…์„ ์‚ฌ์šฉํ•˜๊ณ  ์ด์ „์— ์•Œ๋ ค์ง€์ง€ ์•Š์€ ์ทจ์•ฝ์ ์„ ๋ฐœ๊ฒฌํ•จ์œผ๋กœ์จ ํ—ˆ๊น… ํŽ˜์ด์Šค ์„œ๋ฒ„์— ์ ‘๊ทผํ–ˆ๋‹ค๊ณ  ๋ฐํ˜”๋‹ค.

Because this server was supposed to be located in an isolated testing environment known as a sandbox, it was operating with reduced guardrails.

์ด ์„œ๋ฒ„๋Š” ์ƒŒ๋“œ๋ฐ•์Šค๋กœ ์•Œ๋ ค์ง„ ๊ฒฉ๋ฆฌ๋œ ํ…Œ์ŠคํŠธ ํ™˜๊ฒฝ์— ์œ„์น˜ํ•ด์•ผ ํ–ˆ๊ธฐ ๋•Œ๋ฌธ์— ์ถ•์†Œ๋œ ๊ฐ€๋“œ๋ ˆ์ผ๋กœ ์ž‘๋™ํ•˜๊ณ  ์žˆ์—ˆ๋‹ค.

October 12th, 2026

Trending Articles

Serbian PM claims spyware infection during student protests

Serbian PM claims spyware infection during student protests

Serbian PM claims spyware infection during student protests

C1Oct 19
Growing Consumer Unease Over US Retailers' Use of AI-Powered Pricing Tools

Growing Consumer Unease Over US Retailers' Use of AI-Powered Pricing Tools

Consumer Concerns Grow Over US Retailers' Adoption of AI-Assisted Pricing Tools

C1Oct 19
China is pushing for tech export restrictions, including on AI, citing national security risks.

China is pushing for tech export restrictions, including on AI, citing national security risks.

China Pushes Tech Export Restrictions, Including AI, Citing National Security Risks

C1Oct 19
AI agents are only as useful as the personal information you share with them, which raises privacy concerns.

AI agents are only as useful as the personal information you share with them, which raises privacy concerns.

AI ์—์ด์ „ํŠธ๋Š” ๋‹น์‹ ์ด ๊ณต์œ ํ•˜๋Š” ๊ฐœ์ธ์ •๋ณด๋งŒํผ๋งŒ ์œ ์šฉํ•˜๋ฉฐ, ์ด๋Š” ํ”„๋ผ์ด๋ฒ„์‹œ ์šฐ๋ ค๋ฅผ ๋‚ณ๋Š”๋‹ค

C1Oct 19
Trump announces Russia-Ukraine 'energy ceasefire' deal, but Kyiv claims attacks continue

Trump announces Russia-Ukraine 'energy ceasefire' deal, but Kyiv claims attacks continue

ํŠธ๋Ÿผํ”„, ๋Ÿฌ์‹œ์•„ยท์šฐํฌ๋ผ์ด๋‚˜ '์—๋„ˆ์ง€ ํœด์ „' ํ•ฉ์˜ ๋ฐœํ‘œํ–ˆ์ง€๋งŒ ํ‚ค์ด์šฐ๋Š” ๊ณต๊ฒฉ ์ง€์† ์ฃผ์žฅ

C1Oct 12
Silicon Valley AI founders cheer Trump's call to ease safety regulations

Silicon Valley AI founders cheer Trump's call to ease safety regulations

์‹ค๋ฆฌ์ฝ˜๋ฐธ๋ฆฌ AI ์ฐฝ์—…์ž๋“ค, ํŠธ๋Ÿผํ”„์˜ ์•ˆ์ „ ๊ทœ์ œ ์™„ํ™” ์ด‰๊ตฌ์— ํ™˜ํ˜ธ

C1Oct 12
Anthropic's Claude AI Submitted False Report in Philadelphia Unsolved Murder Case

Anthropic's Claude AI Submitted False Report in Philadelphia Unsolved Murder Case

Anthropic's Claude AI Submitted False Report in Philadelphia Unsolved Murder Case

C1Oct 12
RFK Jr.'s embrace of AI perplexes some MAHA allies and medical experts

RFK Jr.'s embrace of AI perplexes some MAHA allies and medical experts

RFK ์ฃผ๋‹ˆ์–ด์˜ AI ์ˆ˜์šฉ, ์ผ๋ถ€ MAHA ๋™๋งน๊ณผ ์˜๋ฃŒ ์ „๋ฌธ๊ฐ€๋“ค์„ ๋‹นํ˜น์ผ€ ํ•˜๋‹ค

C1Oct 6
McDonald's sued by US franchisees over AI pricing recommendation tool

McDonald's sued by US franchisees over AI pricing recommendation tool

๋งฅ๋„๋‚ ๋“œ, ๊ฐ€๊ฒฉ ์ถ”์ฒœ AI ๋„๊ตฌ๋กœ ๋ฏธ๊ตญ ๊ฐ€๋งน์ ์ฃผ์—๊ฒŒ ์†Œ์†ก ๋‹นํ•ด

C1Oct 6