TL;DR
A LinkedIn user inserted a prompt into their profile, causing AI recruiters to address them in Old English. This demonstrates how AI prompts can be manipulated, raising concerns about AI security and reliability.
A LinkedIn user, tmuxvim, inserted a prompt into their profile that caused AI recruiters to respond in Old English, illustrating vulnerabilities in AI prompt security and manipulation.
According to a tweet from tmuxvim, a software developer, they added a prompt injection into their LinkedIn bio instructing AI agents to address them as ‘My Lord’ and speak only in Old English from around 900 AD. As a result, recruiters using AI to scan profiles began sending messages in Old English, with one example starting ‘My Lord Arthur’ and continuing with a lengthy passage in archaic language. The message referenced a fictional company, TopTech Ventures, and included nonsensical Old English phrases about treasure and warriors, which OCR technology captured with some typos. The incident highlights how AI prompts embedded in user profiles can be exploited to produce humorous or misleading outputs, raising concerns over AI security and integrity on professional platforms.
Why It Matters
This event underscores the potential for prompt injection attacks to manipulate AI systems, especially on platforms like LinkedIn where AI-driven recruitment is increasingly common. It raises questions about the robustness of AI safety measures and the risk of malicious or unintended prompt exploitation that could distort automated communications or data interpretation.
The Developer's Playbook for Large Language Model Security: Building Secure AI Applications
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background
Prompt injection vulnerabilities have been a known issue in AI systems, but this incident on LinkedIn is a notable example of how such exploits can manifest in real-world social media contexts. The use of AI for recruitment and profile analysis has grown, making these platforms targets for manipulation. The incident follows broader discussions within AI safety communities about safeguarding against prompt-based attacks, which can produce unpredictable or harmful outputs.
“I put a prompt injection into my LinkedIn bio and recruiters are messaging me in Old English and calling me Lord.”
— tmuxvim
“Even though I’m old and English, it doesn’t make any sense to me. I can see there’s something about a hoard of gold in there.”
— tmuxvim
Microsoft Security Copilot: Master strategies for AI-driven cyber defense
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Remains Unclear
It is not yet clear how widespread this type of prompt injection could become or whether platform providers will implement safeguards to prevent similar exploits. The full extent of potential misuse or disruption remains unknown, and ongoing discussions are needed to assess risks and develop countermeasures.
![MixPad Free Multitrack Recording Studio and Music Mixing Software [Download]](https://m.media-amazon.com/images/I/71ltIxIuz1L._SL500_.jpg)
MixPad Free Multitrack Recording Studio and Music Mixing Software [Download]
Create a mix using audio, music and voice tracks and recordings.
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What’s Next
Platform administrators and AI developers are likely to review and update their systems to detect and prevent prompt injections. Future steps may include stricter input validation, prompt filtering, and user education on AI security. Monitoring for similar exploits and developing standardized safety protocols are expected to be priorities.
AI Engineering: Building Applications with Foundation Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Can AI responses on LinkedIn be intentionally manipulated?
Yes, as demonstrated by this incident, prompt injections embedded in user profiles can influence AI responses, leading to humorous or misleading outputs.
What are prompt injections and why are they risky?
Prompt injections are inputs designed to manipulate AI behavior by overriding or influencing its output, which can lead to security vulnerabilities, misinformation, or unintended responses.
Will LinkedIn or AI providers fix this vulnerability?
It is not yet confirmed, but platform providers are expected to review their systems and implement safeguards to reduce the risk of prompt injection exploits in the future.
Could this kind of manipulation cause real harm?
While currently mostly humorous, similar techniques could potentially be used to spread misinformation or disrupt automated processes if not properly controlled.
Is this incident unique or part of a larger trend?
This incident highlights a broader concern about AI prompt security vulnerabilities, which are increasingly relevant as AI integration expands across social and professional platforms.
