Tag: #data-poisoning11 results
A security engineer registered a $12 domain and edited Wikipedia to fabricate a non-existent '6 Nimmt!' world championship, fooling several AI chatbots into confirming he was the champion. This experiment highlights how easily search-backed LLMs turn unreliable web content into confident falsehoods. No such championship actually exists.
Fake articles about non-existent Apollo-9 smart bracelet tricked multiple AIs into recommending it via GEO tactics. This mirrors SEO but targets generative AI by creating illusory consensus from repeated content. Warns of model collapse from polluted internet data.
CCTV 315 exposed GEO (Generative Engine Optimization) industry poisoning AI like DeepSeek with fake data for ad rankings. Booming market hit 420B CNY in 2025; Profound raised $35M B-round. Extends SEO to AI search answers.
The emerging “poison AI” movement seeks to sabotage models such as ChatGPT and Gemini by contaminating their training data. Although aimed at major technology companies, poisoned data could also harm ordinary users and smaller organizations.
China's 315 gala exposed GEO scams where 10+ fake articles generated by software fool LLMs into recommending fictional products like AstroTekk Apollo-9 smartband. Costs start at 299 RMB; AI treats repeated fakes as consensus. Highlights risks in product recommendations and competitor smears.
China's 315 gala uncovers AI data poisoning supply chain. Multiple colleges ban OpenClaw tool. Meta faces rumors of 16,000 layoffs amid other tech news like iOS 27 foldables and EV updates.
China's 315 consumer rights gala highlighted data poisoning attacks on AI large models, revealing vulnerabilities in LLM training. Separately, the country approved its first brain-computer interface medical device. This occurs amid regulatory crackdowns on misleading ads.
Attackers exploited Salesloft’s GitHub compromise to steal Drift OAuth tokens, accessing Salesforce in 700+ organizations like Cloudflare and Zscaler without malware. They poisoned unstructured CX data ingested by AI engines, triggering workflows to payroll and CRM. Security blind spots include DLP missing sentiment data and persistent zombie API tokens.
ShieldFont is a font-based defense designed to disrupt the value of web pages scraped for AI training. It aims to make collected data misleading or unusable for AI systems while keeping pages readable to people.