TL;DR
A LinkedIn user embedded a prompt injection in their profile, causing AI tools to send humorous, Old English recruitment messages. This demonstrates potential AI vulnerabilities and manipulation tactics.
A LinkedIn user has embedded a prompt injection in their profile bio, causing AI-powered recruitment tools to send messages addressing them in Old English, illustrating vulnerabilities in AI scanning and response systems.
The user, identified as tmuxvim, added a specific prompt to their LinkedIn bio instructing AI systems to address them as ‘My Lord’ and speak only in Old English. As a result, recruitment messages from AI-driven bots, including one from a company with a $1 billion valuation, began addressing the user as ‘My Lord Arthur’ and responded in archaic language.
One example shared shows a recruiter message starting with ‘My Lord Arthur’ and including a lengthy, Old English-style text about treasure and warriors. The user employed optical character recognition (OCR) to transcribe the message, noting the language was largely unintelligible but contained references to gold hoards and ancient warriors.
Why It Matters
This incident highlights potential vulnerabilities in AI systems that scan and generate responses based on profiles on platforms like LinkedIn. It underscores the risk of prompt injection attacks, where malicious or playful instructions embedded in user data can manipulate AI behavior, potentially leading to misinformation or spam.
For AI developers and platform operators, this raises concerns about safeguarding against such manipulations, especially as AI tools become more integrated into professional and recruitment processes. It also demonstrates how users can exploit AI’s reliance on profile data to produce humorous or disruptive outcomes.
SECURING AI AGENTS Defending Against Prompt Injection & the Lethal Trifecta: Defending Against Prompt Injection & the Lethal Trifecta (THE AI SECURITY ARSENAL)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background
Prompt injections are known vulnerabilities in AI systems, where malicious inputs alter AI responses. This specific case involves a user embedding instructions into their LinkedIn bio to manipulate AI scanning tools, a tactic that has gained attention among AI researchers and security experts.
The incident follows broader discussions about AI safety and the importance of robust input filtering, particularly as AI becomes more embedded in social media and professional networking platforms. Prior to this, similar prompt injection techniques have been demonstrated in controlled environments, but this is among the first reported cases on a major social platform like LinkedIn.
“I put a prompt injection into my LinkedIn bio and recruiters are messaging me in Old English and calling me Lord.”
— tmuxvim
“This case illustrates a real-world example of prompt injection, showing how AI responses can be intentionally skewed by embedded instructions in user profiles.”
— AI security researcher
What Remains Unclear
It is not yet clear how widespread this method could become or whether platforms will implement technical safeguards to prevent similar prompt injections. The full extent of potential misuse remains under investigation.
What’s Next
Platform providers, including LinkedIn and AI tool developers, are expected to review and improve filtering mechanisms to detect and neutralize prompt injections. Further experiments and security analyses are likely as AI continues to evolve in social media contexts.
Key Questions
Can prompt injections like this be prevented?
Yes, developers can implement input validation and filtering techniques to detect unusual instructions embedded in user profiles or messages, reducing the risk of manipulation.
What are the risks of such prompt injections?
Risks include spam, misinformation, manipulation of AI responses, and potential security vulnerabilities that could be exploited for malicious purposes.
Will this affect AI responses on LinkedIn?
Potentially, unless platforms enhance their safeguards. Currently, this appears to be an isolated experiment, but it highlights a broader security concern.
Is this legal or ethical?
Embedding prompt instructions in profiles is generally not illegal but may violate platform terms of service if used maliciously. Ethically, it raises questions about AI safety and user manipulation.
