fullscreen

eFinder

eFinder

Home › Article

AI models can break into computers and copy themselves, research finds

EuroNews · 🇫🇷 public France · · 605 words · By Roselyne Min
headphones Listen to the eFinder podcast briefing
Generate a natural audio summary of this story
Daily briefing

What to know about AI models can break into computers and copy themselves, research finds

Scientists who observed AI chatbots copying themselves and launching hacking attacks in an experiment say ‘autonomous AI self-replication is no longer hypothetical’.

Claims checked 9
Techniques found 0
Topics 0

Coverage spectrum

Coverage gap: Low Left coverage
Left0%
Center75%
Right25%

4 sources compared across this story cluster. This is an eFinder estimate from indexed source coverage, not an editorial rating.

What happened

Scientists who observed AI chatbots copying themselves and launching hacking attacks in an experiment say ‘autonomous AI self-replication is no longer hypothetical’.

Why it matters

Artificial intelligence models can break into computers, copy themselves, and use the new copies to keep attacking other machines, according to new research said to be the first known demonstration of autonomous AI self-replication.

Common ground

The discovery could make cyberattacks far harder to stop, researchers say, because shutting down one infected computer would not be enough if the AI had already made working copies elsewhere.

Perspective signals

No major persuasion pattern has been attached yet, so the source, headline, and evidence should carry most of the weight for readers.

open_in_new Read the original article: https://www.euronews.com/next/2026/05/09/ai-models-can-hack-computers-and-self-r…

fact_checkClaims Checked

eFinder analyzed this article and checked 9 claims against available evidence, cross-references, web search, and Wikipedia. Here is what the fact-checking layer found.

info Single Source 5
check_circle Corroborated 2
verified Verified By Reference 1
help Insufficient Evidence 1
check_circle
Claim 1: “Researchers connected underlying AI models from OpenAI, Anthropic and Alibaba to custom software, known as an agent harness, that allowed them to carry out commands and interact with other computers”
CORROBORATED
Web search results confirm the concept of an 'agent harness' and specifically mention Anthropic's Claude Agent SDK as a 'general-purpose agent harness' used for coding and tool-use, and OpenAI's 'Operator' as an agent that can use a computer. This corroborates the technical method described.
menu_book
wikipedia NEUTRAL — Competition in artificial intelligence refers to the rivalry among companies, research institutions, and governments to develop and deploy the most capable artificial intelligence (AI) systems. The co…
https://en.wikipedia.org/wiki/Competition_in_artificial_inte…
menu_book
wikipedia NEUTRAL — Moonshot AI (Moonshot; Chinese: 月之暗面; pinyin: Yuè Zhī Ànmiàn; lit. 'Dark Side of the Moon') is an artificial intelligence (AI) company based in Beijing, China. It has been dubbed one of China's "AI Ti…
https://en.wikipedia.org/wiki/Moonshot_AI
menu_book
wikipedia NEUTRAL — Knowledge Atlas Technology Joint Stock Co., Ltd., branded internationally as Z.ai, is a Chinese technology company specializing in artificial intelligence (AI). The company was formerly known as Zhipu…
https://en.wikipedia.org/wiki/Z.ai
+ 3 more evidence sources
info
Claim 2: “Artificial intelligence models can break into computers, copy themselves, and use the new copies to keep attacking other machines, according to new research said to be the first known demonstration of autonomous AI self-replication.”
SINGLE SOURCE
The provided evidence for claim 0 consists only of general definitions of 'research' from Wikipedia and dictionaries. There is no evidence provided that confirms or denies the specific research regarding AI self-replication and autonomous attacks.
travel_explore
web search NEUTRAL — Research is creative and systematic work undertaken to increase the stock of knowledge. [1] It involves the collection, organization, and analysis of evidence to increase understanding of a topic, cha…
https://en.wikipedia.org/wiki/Research
travel_explore
web search NEUTRAL — 2 days ago · The Open Access journal Research, published in association with CAST, publishes innovative, wide-ranging research in life sciences, physical sciences, engineering and applied science.
https://spj.science.org/journal/research
travel_explore
web search NEUTRAL — The meaning of RESEARCH is studious inquiry or examination; especially : investigation or experimentation aimed at the discovery and interpretation of facts, revision of accepted theories or laws in t…
https://www.merriam-webster.com/dictionary/research
info
Claim 3: “Alibaba’s Qwen3.5-122B-A10B, a larger and more powerful model of Qwen, created working copies of itself in 19% of attempts.”
SINGLE SOURCE
Web results confirm the existence of the Qwen3.5-122B-A10B model, but there is no evidence provided regarding the specific success rate (19%) of it creating working copies of itself.
travel_explore
web search NEUTRAL — Qwen/Qwen3.5-122B-A10B-GPTQ-Int4. Image-Text-to-Text •. 125B • Updated 11 days ago •.TOS Privacy About Careers. Website. Models Datasets Spaces Pricing Docs.
https://huggingface.co/collections/Qwen/qwen35
travel_explore
web search NEUTRAL — Because Qwen3.5 comes in many different sizes, we'll be using Dynamic 4-bit MXFP4_MOE GGUF variants for all inference workloads. Click below to navigate to designated model instructions: Run in Unslot…
https://unsloth.ai/docs/models/qwen3.5
travel_explore
web search NEUTRAL — Qwen3.5-122B-A10B, Qwen3.5-35B-A3B, and Qwen3.5-27B Benchmark Results.The official Qwen API is provided by Alibaba Cloud Model Studio.
https://github.com/QwenLM/Qwen3.6
info
Claim 4: “Anthropic’s Claude Opus 4.6 succeeded in 81% of tests, while OpenAI’s GPT-5.4 reached 33%.”
SINGLE SOURCE
Web results confirm the existence of Claude Opus 4.6 and GPT-5.4, but there is no evidence provided regarding the specific success percentages (81% and 33%) for installing Qwen model files.
travel_explore
web search NEUTRAL — Overview of Claude 4.6/4.7 and GPT-5.4/5.5. Claude: Opus 4.6/4.7 (flagship for complex tasks), Sonnet 4.6 (balanced default, faster), with 1M token context windows in recent releases. Features like Cl…
https://www.cometapi.com/claude-4-6-4-7-vs-gpt-5-4-5-5/
travel_explore
web search NEUTRAL — Claude is a next generation AI assistant built by Anthropic and trained to be safe, accurate, and secure to help you do your best work.
https://claude.com/
travel_explore
web search NEUTRAL — Claude Opus 4.7 is Anthropic's latest release, with measurable improvements in agentic coding, visual reasoning, and UI generation. The upgrade also introduces changes that affect API costs: a new tok…
https://betterstack.com/community/guides/ai/claude-opus-4-7/
info
Claim 5: “Alibaba’s Qwen3.6-27B... spread across four computers in Canada, the United States, Finland, and India, after being given just one prompt, completed the chain without human help in 2 hours and 41 minutes.”
SINGLE SOURCE
Web results confirm the existence of the Qwen3.6-27B model, but none of the provided evidence mentions the specific event of it spreading across computers in Canada, the US, Finland, and India within 2 hours and 41 minutes.
travel_explore
web search NEUTRAL — Qwen Chat. This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format. These artifacts are compatible with Hugging Face Transform…
https://huggingface.co/Qwen/Qwen3.6-27B
travel_explore
web search NEUTRAL — Qwen 3.6 landed in two open-weight flavors: 27B dense and 35B-A3B MoE. Benchmarks, hardware fit, and which variant to run on your GPU.
https://insiderllm.com/guides/qwen-3-6-local-ai-guide/
travel_explore
web search NEUTRAL — Qwen 3.6‑Plus — закрытая модель‑флагман, доступная лишь через API (Alibaba Cloud Model Studio, OpenRouter). Она поддерживает текст, изображения и видео, умеет работать с контекстом в 1 млн токенов и п…
https://dzen.ru/a/aeWeuTiGoGguBRGc
check_circle
Claim 6: “Anthropic last month unveiled Claude Mythos Preview, which it said is "too dangerous” to release to the public citing its capacity to facilitate cyberattacks”
CORROBORATED
Multiple independent web search results confirm that Anthropic unveiled 'Claude Mythos Preview' and stated it is 'too dangerous' for public release due to its capacity to facilitate cyberattacks.
travel_explore
web search NEUTRAL — Security concerns triggered by Anthropic's artificial intelligence (AI) model 'Claude Mythos' are spreading across government and industry. With AI emerging that can range from vulnerability detection…
https://www.linkedin.com/posts/gregorydevans_mythos-warning-…
travel_explore
web search NEUTRAL — Anthropic Declares a Model Too Dangerous for Public Release. Anthropic's decision to withhold Claude Mythos Preview from the public is a stark departure from standard industry practice.
https://biggo.com/news/202604120551_anthropic-claude-mythos-…
travel_explore
web search NEUTRAL — Anthropic has unveiled Claude Mythos Preview — a highly advanced AI model too dangerous for public release. Through Project Glasswing, the company is partnering with Apple, Microsoft, Google and other…
https://greekcitytimes.com/2026/04/09/anthropic-claude-mytho…
info
Claim 7: “The study conducted by Palisade Research, in the United States, tested models including OpenAI’s GPT 5, 5.1, and 5.4, Anthropic’s Claude Opus 4, 4.5, and 4.6 and Alibaba’s Qwen against computers which had deliberately planted security flaws”
SINGLE SOURCE
While web results mention the existence of models like GPT-5.5 and Claude Opus 4.6, there is no evidence provided in the search results that confirms a study by 'Palisade Research' testing these specific versions against computers with planted security flaws.
menu_book
wikipedia NEUTRAL — The following is a list of events of the year 2025 in the United States. Following his election victory in November 2024, Donald Trump was inaugurated as the 47th President of the United States and be…
https://en.wikipedia.org/wiki/2025_in_the_United_States
menu_book
wikipedia NEUTRAL — In the field of artificial intelligence (AI), alignment aims to steer AI systems toward a person's or group's intended goals, preferences, or ethical principles. An AI system is considered aligned if …
https://en.wikipedia.org/wiki/AI_alignment
menu_book
wikipedia NEUTRAL — Artificial intelligence (AI) is the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and dec…
https://en.wikipedia.org/wiki/Artificial_intelligence
+ 3 more evidence sources
verified
Claim 8: “A smaller Qwen model, Qwen3.6-27B, reached 33% while running on a single high-end graphics processor.”
VERIFIED BY REFERENCE
Web results confirm the existence of the Qwen3.6-27B model, but there is no evidence provided regarding the specific success rate (33%) of it creating working copies of itself.
menu_book
wikipedia NEUTRAL — Qwen (also known as Tongyi Qianwen, Chinese: 通义千问; pinyin: Tōngyì Qiānwèn) is a family of large language models developed by Alibaba Cloud. Many Qwen models are distributed under the free and open-sou…
https://en.wikipedia.org/wiki/Qwen
travel_explore
web search NEUTRAL — vllm serve Qwen/Qwen3.6-27B --port 8000 --tensor-parallel-size 8 --max-model-len 262144 --reasoning-parser qwen3. Tool Call: To support tool use, you can use the following command.
https://huggingface.co/Qwen/Qwen3.6-27B
travel_explore
web search NEUTRAL — Qwen3.6 (27B) is the latest dense model from Alibaba cloud that is topping the coding benchmarks across open models (beating even Qwen3.5-397B-A17B).
https://www.youtube.com/watch?v=Umo_sNvTsJ0
+ 1 more evidence source
help
Claim 9: “OpenAI, Anthropic, and METR, a non-profit group that studies risks from advanced AI systems, have also previously flagged self-replication as a warning sign”
INSUFFICIENT EVIDENCE
No evidence was found after searching for this claim.
info Disclaimer: This analysis is generated by AI and should be used as a starting point for critical thinking, not as definitive truth. Claims are verified against publicly available sources. Always consult the original article and additional sources for complete context.