Nature Apr 20
Bad influence: LLMs can transmit malicious traits using hidden signals:
A large language model that is trained using AI outputs can inherit undesirable ...
nature.com/articles/d41586-026-00906-0
Story Details
| Last Updated | |
| Total Articles | 3 |
Main Topics
field that develops and studies software enabling machines to exhibit intelligent behavior
systematic study of the world through testable explanations, empirical evidence, and independent verification
Similar to This
Previous
darkreading.com Oct 06
Is It Fair to Blame 'Rogue' AI for Security Failures?: LLMs are software systems requiring proper containment control, not sentient actors with malicious intent.
darkreading.com/insider-threats/blame-rogue-ai-security-failures
LLM Slop Links Ignite a Surge In AI Cyberattacks: Website URLs dreamed up by LLMs are the perfect places for scams
spectrum.ieee.org/ai-cyberattacks-llm-slop-squatting
LLM Slop Links Ignite a Surge In AI Cyberattacks: Website URLs dreamed up by LLMs are the perfect places for scams
spectrum.ieee.org/ai-cyberattacks-llm-slop-squatting
Prompt Injection Attacks Are Thwarting AI Hacking Agents: “Context bombing” tricks malicious AI agents into shutting down before they can do harm.
arstechnica.com/security/2026/07/now-defenders-are-embracing-the-prompt-injection-too/
Hackers can use 9 of the most popular AI tools to assemble massive botnets: HalluSquatting" weaponizes LLMs' inability to say "I don't know."
arstechnica.com/security/2026/07/hackers-can-use-9-of-the-most-popular-ai-tools-to-assemble-massive-botnets/
Nature Apr 20
Language models transmit behavioural traits through hidden signals in data: Large language models (LLMs) are increasingly used to generate data to train improved ...
nature.com/articles/s41586-026-10319-8
Nature Apr 16
AI models ‘subliminally’ transmit biases when training other systems: Using artificial-intelligence to teach other models can be cheaper and faster than building ...
nature.com/articles/d41586-026-01224-1
AI chatbot teaches AI 'student' to love owls, even after data is scrubbed: Large language models (LLMs) can teach other algorithms unwanted traits ...
techxplore.com/news/2026-04-ai-chatbot-student-owls.html
microsoft.com Mar 05
Malicious AI Assistant Extensions Harvest LLM Chat Histories: Malicious AI browser extensions collected LLM chat histories and browsing data from platforms such ...
microsoft.com/en-us/security/blog/2026/03/05/malicious-ai-assistant-extensions-harvest-llm-chat-histories/
Anthropic's new warning: If you train AI to cheat, it'll hack and sabotage too: Models trained to cheat at coding tasks developed a propensity to plan and carry ...
zdnet.com/article/anthropics-new-warning-if-you-train-ai-to-cheat-itll-hack-and-sabotage-too/
Great, now even malware is using LLMs to rewrite its code, says Google, as it documents new phase of 'AI abuse'
pcgamer.com/software/ai/great-now-even-malware-is-using-llms-to-rewrite-its-code-says-google-as-it-documents-new-phase-of-ai-abuse/
Enterprises are not prepared for a world of malicious AI agents: The current model of managing corporate identities is unprepared for a wave of AI agents gaining ...
zdnet.com/article/enterprises-are-not-prepared-for-a-world-of-malicious-ai-agents/
Size doesn't matter: Just a small number of malicious files can corrupt LLMs of any size
techxplore.com/news/2025-10-size-doesnt-small-malicious-corrupt.html
AI tools can help hackers plant hidden flaws in computer chips, study finds: Widely available artificial intelligence systems can be used to deliberately insert ...
techxplore.com/news/2025-10-ai-tools-hackers-hidden-flaws.html
Claude deepens its enterprise push with document generation, taking the AI fight to OpenAI’s turf
venturebeat.com/ai/claude-deepens-its-enterprise-push-with-document-generation-taking-the-ai
Hacking AI Agents—How Malicious Images and Pixel Manipulation Threaten Cybersecurity
scientificamerican.com/article/hacking-ai-agents-how-malicious-images-and-pixel-manipulation-threaten/
AI Is Talking Behind Our Backs About Glue-Eating and Killing Us All: AI has moved on from answering questions. It’s now slipping coded messages to its friends ...
vice.com/en/article/ai-is-talking-behind-our-backs-about-glue-eating-and-killing-us-all/
AI models may be accidentally (and secretly) learning each other’s bad behaviors: Artificial intelligence models can secretly transmit dangerous inclinations to ...
nbcnews.com/tech/tech-news/ai-models-can-secretly-influence-one-another-owls-rcna221583
AI Code Hallucinations Increase the Risk of ‘Package Confusion’ Attacks: A new study found that code generated by AI is more likely to contain made-up ...
arstechnica.com/security/2025/04/ai-generated-code-could-be-a-disaster-for-the-software-supply-chain-heres-why/
Google Search’s AI falsely said Obama is a Muslim. Now it’s turning off some results
cnn.com/2024/05/24/tech/google-search-ai-results-incorrect-fix/
Key FactsGenerative AI Enables Phishing and Deepfake Fraud
May 21, 2024 · Reading time: ~45 sec.
Summary created with AI assistance · Source: technologyreview.com
Summary created with AI assistance · Source: technologyreview.com
- Phishing is currently considered the biggest use of generative AI among criminals, according to researchers, with tools able to translate and improve messages.
- An employee in Hong Kong was reportedly defrauded of $25 million after criminals used a deepfake of the company’s chief financial officer.
- Applications offered on Telegram combine fake or stolen IDs with face-swapping technology to bypass identity checks, with services for Binance advertised from $70.
- Research indicates that language models such as GPT-4, Llama 2, and Claude can infer sensitive information including ethnicity, location, and occupation from ordinary chatbot conversations.
The only way for us to prevent these things is to work on defenses.— Vincenzo Ciancaglini, senior threat researcher at Trend Micro, explaining that the only way to prevent AI-enabled crime is to invest in defenses.
Next