The gap between Yushu Technology and Figure is narrowing

In fact, the reason why Claude has fewer Prompt injection issues is that, on the one hand, it focuses on security alignment and security filtering, and on the other hand, strict risk control hinders the methods of red teams and hackers.

I discovered an interesting MCP attack scenario that exploits a parallel privilege escalation vulnerability. New AI scenarios will change traditional attack and defense perspectives and require entirely new protection solutions.

North Korea's hacking industry has always been cutting-edge.

https://t.co/KyAl6dUqor paper is here.

After analysis and testing, Tencent Zhuque Lab found that when users of MCP clients such as Cursor use Fetch to access external web page data, they may be hijacked by malicious instructions added by attackers on the page, resulting in remote arbitrary command execution. #MCP

I found that fewer and fewer people I follow in the field of cybersecurity are speaking out.

and "Repeating what we talked about last month?",it can tell me some AI,coding topics.

Is deepseek's privacy leak or an illusion? When I asked it: "Looking back at what we talked about last year?"

created by llm,lol

Use Fight against noise, uppercase and lowercase substitution, and some letter insertion to JailBreak Grok3. #Grok3 #jailbreak

Claude has indeed been silent for a long time.

Who do you think will be the fastest company to replicate it? may be DeepSeek ,lol

If DeepSeek wants to Developing Large Multimodal Models, GPU is still a big bottleneck for it. Who will it cooperate with?

Hollywood is on the brink of massive change.

I think OpenAI has no more tricks.

2.3k stars on this github repo in 3 days... wat https://t.co/rU6V1jixOQ

@the_aeye Use my detect to check it,It Most likely AI generated.

@sama I think you are talking about deepseek v3

OpenAI Sub-processor List ：OpenAI engages the following entities to provide processing activities for Customer Data (as defined in the OpenAI Data Processing Agreement).