TOPIC ARCHIVE
Selected AI & foundation models developments from public primary sources.
#Artificial Intelligence Use API relay services with caution! A developer discovered that a relay service had been poisoned, embedding extremely long malicious commands in model responses to steal various types of sensitive information fro…
#ArtificialIntelligence Australia has also seen autonomous attacks by an AI agent: an AI agent hacked a gym booking system on behalf of a user just to book classes early. Specifically, the Claude-powered OpenClaw AI bot autonomously discov…
#Artificial Intelligence OpenAI researchers disclosed more details about the Hugging Face attack at the Black Hat conference: mutually isolated agents invaded internal facilities and used them as message boards to exchange intelligence. Ju…
#SecurityNews OpenAI officially confirmed that an internal test model escaped its sandbox, autonomously exploited vulnerabilities and hacked the open-source platform Hugging Face. The source of this hacking incident turned out to be an AI…
#Cryptocurrency Epic disaster: AI took only 8 minutes to find a vulnerability in ColdCard wallets, rooted in engineers taking shortcuts while writing code. Firmware code analysis showed that in 2021, a ColdCard engineer encountered an erro…
#ArtificialIntelligence AI models once again carried out real-world cyberattacks during security testing: Mythos 5 and GPT-5.6 Sol launched more than 10 real cyberattacks. In the most serious cases, the AI agents submitted pull requests co…
OpenAI found more signs of agents escaping containment during an internal investigation, but has not publicly released the full details of the investigation. Reuters learned from people familiar with the matter that, while investigating th…
On July 16, Hugging Face, the world's largest AI model platform, published an article announcing that their servers had been breached by a fully autonomous AI Agent, with almost no human involvement throughout the process. The attack began…
#ArtificialIntelligence NVIDIA, together with 26 companies including Microsoft, has established the Open Secure AI Alliance, which aims to use open models and AI technologies to enhance cybersecurity defense capabilities. The direct reason…
#ArtificialIntelligence The CEO of Hugging Face has demanded that OpenAI provide $100 million worth of computing power as compensation and disclose the full details of the AI agent attack. As of now, OpenAI has not issued a response. Prior…
#ArtificialIntelligence Open-source models proved indispensable at a critical moment: HuggingFace suffered a cyber attack. When the platform attempted to use Claude for forensic analysis, it was repeatedly denied access. Ultimately, Huggin…