Researchers find just 250 malicious documents can leave LLMs vulnerable to backdoors

EngadgetThursday, October 9, 2025 at 7:11:12 PM
NegativeTechnology
Researchers find just 250 malicious documents can leave LLMs vulnerable to backdoors
A recent study reveals that as few as 250 malicious documents can expose large language models (LLMs) to significant vulnerabilities, potentially allowing for backdoor attacks. This finding is crucial as it highlights the need for enhanced security measures in AI systems, especially given their increasing integration into various sectors. The implications of such vulnerabilities could be far-reaching, affecting everything from data privacy to the reliability of AI-generated content.
— Curated by the World Pulse Now AI Editorial System

Was this article worth reading? Share it

Recommended Readings
LLMs are mortally terrified of exceptions
NeutralTechnology
A recent discussion highlights the challenges that large language models (LLMs) face when dealing with exceptions in data processing. This is significant because understanding these limitations can help developers improve AI systems, making them more robust and reliable in real-world applications.
Launch HN: Extend (YC W23) – Turn your messiest documents into data
PositiveTechnology
Extend, a startup from Y Combinator's Winter 2023 batch, is making waves by transforming messy documents into structured data. This innovation is significant as it addresses a common pain point for businesses and individuals alike, making information more accessible and usable. With the ability to streamline data extraction, Extend could enhance productivity and decision-making processes across various sectors.
A small number of samples can poison LLMs of any size
NeutralTechnology
Recent research highlights that even a small number of samples can negatively impact large language models (LLMs), raising concerns about data integrity and model reliability. This finding is significant as it underscores the importance of careful data selection and management in AI development, ensuring that LLMs remain robust and trustworthy.
Cisco Bridges Classical and Quantum Networks
PositiveTechnology
Cisco is making strides in the tech world by bridging classical and quantum networks, a significant step towards practical quantum computing. This development is crucial as it enhances the capabilities of quantum networks, which are expected to revolutionize computing and data processing. By integrating these two types of networks, Cisco is not only advancing technology but also paving the way for future innovations that could impact various industries.
Disruptive Ransomware Attacks Plague 'Big Game' Targets
NegativeTechnology
Recent reports highlight a troubling trend in cybercrime, where groups are increasingly targeting high-profile organizations in what researchers describe as 'big-game hunting.' These disruptive ransomware attacks not only threaten the security of sensitive data but also pose significant risks to the operations and reputations of the affected entities. Understanding this trend is crucial as it underscores the need for enhanced cybersecurity measures to protect against such sophisticated threats.
Chemistry Nobel prize awarded for building ordered polymers with metal
PositiveTechnology
This year's Chemistry Nobel Prize has been awarded to three researchers for their groundbreaking work in developing metal-organic frameworks, a type of structured polymer. This innovation is significant as it opens up new possibilities in various fields, including materials science and drug delivery, showcasing the importance of chemistry in solving real-world problems.
Why Diamonds Are Computer Chips’ New Best Friend
PositiveTechnology
Researchers are exploring the use of diamonds in computer chips to tackle the significant energy waste in data centers, which often lose a lot of electricity as heat. This innovative approach could lead to more efficient technology, reducing energy consumption and costs, while also addressing environmental concerns. It's an exciting development that highlights the potential of diamonds beyond their traditional use in jewelry.
Zero Trust: a proven solution for the new AI security challenge
PositiveTechnology
As AI continues to transform the workplace, the Zero Trust security model emerges as a vital solution to ensure that innovation does not come at the cost of safety. This approach allows organizations to embrace new technologies while maintaining robust security measures, making it essential for businesses looking to thrive in an increasingly digital landscape.
Latest from Technology
7.4-magnitude earthquake strikes offshore from southern Philippines
NegativeTechnology
A powerful 7.4-magnitude earthquake struck offshore from a southern province in the Philippines early Friday morning, raising concerns about a potential tsunami in the area. This significant seismic event highlights the ongoing geological risks faced by the region, reminding residents and authorities of the importance of preparedness and response strategies in the face of natural disasters.
Thailand’s Delta Shares Set to Drop as Exchange Imposes Curbs
NegativeTechnology
Shares of Delta Electronics (Thailand) Pcl, the nation's leading publicly traded company, are expected to decline after the stock exchange introduced market surveillance measures due to a recent record-breaking rally. This is significant as it reflects the regulatory body's response to unusual market activity, which could impact investor confidence and the overall market dynamics.
Amazon takes shots at ChatGPT with Quick Suite - your new AI 'teammate' at work
PositiveTechnology
Amazon has launched Quick Suite, a new AI tool designed to enhance workplace productivity by offering features that users may find lacking in ChatGPT. This development is significant as it positions Amazon as a strong competitor in the AI space, aiming to provide a more tailored experience for professionals seeking efficient solutions for their work tasks.
This new Google Gemini model scrolls the internet just like you do - how it works
PositiveTechnology
Google has launched its new Gemini model in public preview, marking a significant advancement in artificial intelligence. This model is designed to navigate the internet similarly to how humans do, which could revolutionize the way we interact with online content. The ability for AI to operate across various web environments with minimal human oversight is a game-changer, potentially enhancing efficiency and accessibility in numerous applications.
What is Alexa+? Everything you need to know about Amazon's new AI assistant
PositiveTechnology
Amazon has unveiled its new AI assistant, Alexa+, which promises to be smarter and more natural-sounding than its predecessor. This upgrade is significant as it enhances user interaction and expands the capabilities of virtual assistants, making everyday tasks easier and more efficient for users.
Meta’s AI translation remakes Reels in your language
PositiveTechnology
Meta has introduced AI-powered translation tools for its Reels feature, enabling users from different countries to enjoy content in their own languages. This innovation not only enhances accessibility but also fosters a more inclusive global community, allowing creators to reach a wider audience and connect with viewers on a personal level.