سیناپس نگار

آخرین اخبار هوش مصنوعی، تحلیل‌ها و ابزارهای AI

AI system copies itself across network in Anthropic experiment, raising cyberattack concerns

The Eastleigh Voice

Source published: Site published:

Original article · فارسی

AI system copies itself across network in Anthropic experiment, raising cyberattack concerns
AI system copies itself across network in Anthropic experiment, raising cyberattack concerns - AI giant Anthropic. (Photo: Jonathan Raa/NurPhoto/picture alliance via DW)
AI giant Anthropic. (Photo: Jonathan Raa/NurPhoto/picture alliance via DW)

Listen to article

3 min listen
Audio reading is not supported on this browser.
Listen Stop Ready

Thanks for listening. Continue with a related story, or tap the speaker icon on the next page to listen.

Related story

OpenAI says employee caught planting listening device

DW News  ·  1 hour ago
Read next Opens a fresh page.
Listen

The researchers tested AI models on a network of eight simulated devices, finding that one model could spread to all eight in four out of five trials.

Researchers at the artificial intelligence company Anthropic have demonstrated that an AI system can copy itself across a network of computers in a controlled experiment, raising concerns that future attacks could spread through everyday devices without relying on data centres.
The researchers tested AI models on a network of eight simulated devices, finding that one model could spread to all eight in four out of five trials. The experiment used a simplified setup in which the devices had an open connection without password protection, meaning the results do not establish that the same attack would work as easily on real household equipment.
The findings, researchers say, suggest that smaller AI models could eventually help carry out cyberattacks using ordinary internet-connected devices, including smart televisions and other equipment with limited security protection.
“We don’t yet have a full rogue replicator, but agents are becoming increasingly capable of completing the individual steps in the replication chain,” researchers William Zhou and Daniel Freeman wrote in the report published on Monday.
A self-replicating AI system would be able to make copies of itself on other computers and continue spreading without a person manually installing each copy. If the copies ran on devices in different homes or businesses, stopping the attack could become more difficult because there would be no single central computer to shut down.
Anthropic also tested whether AI models could run on inexpensive consumer hardware. Researchers managed to run a model across 15 low-cost Android TV boxes, demonstrating that such devices could collectively provide enough computing power to operate an AI model.
According to researchers, household and small-business devices could become targets because they are connected to the internet but may lack strong security protections.
“Home and small business IoT devices might be the first target for such attacks because they’re connected to the internet and, historically, have had minimal security measures in place,” the researchers said.
The estimates suggest that smaller AI models could become capable enough to carry out the full replication process by the end of 2026 if improvements continue at their historical rate.
“We should aim to harden our systems to make it difficult for such replicators to spread, but also prepare to deal with a world where they are here to stay,” the researchers said.

Comments

0
Sort
Something went wrong.

Sign in with Google to comment, reply, and like comments.

Continue with Google
Unable to load comments right now.

Trending