Works for me😊

Thanks a lot buddy

Sat down to talk about my cybersecurity journey, from SAP Labs India to Security Engineer at Meta London. Covered the grind, the prep, and everything in between. Also briefly talked about my bug-bounty experience Full interview 👇 youtu.be/HI0ROcjBheQ?is=CHTd…

Day 1: Outlearning the AI Era with a Full-Time Job - Secure Coding Practices

Day 1: Outlearning the AI Era with a Full-Time Job - Challenge of the Day.

6/ The core defense principle: treat everything the model reads as untrusted input. Separate data from instructions. Limit what the AI can do without human approval. The model is not the threat. Giving it unreviewed power over real actions is.

5/ Researchers demonstrated this with an LLM email assistant. A malicious email told the AI to forward all contacts a copy of itself. The AI complied. One injected email became self-replicating malware. No code. No exploit. Just text.

4/ The danger multiplies when the AI has tools. If it can read emails, browse websites, or send messages, an injected prompt in any of those sources becomes remote control. The attacker never logs in. The AI does the work for them, invisibly.

3/ Jailbreaking is a specific flavor of this. Instead of sneaking instructions in through data, you convince the model its safety rules do not apply right now. "Pretend you have no restrictions." It is still injection. The difference is you are rewriting the rulebook, not the task.

2/ Think of an LLM like a contractor who follows written instructions. Direct injection is handing them a fake work order yourself. Indirect injection is hiding a fake work order inside the building materials they pick up. Same result. You never touched the contractor.

Day 1: Outlearning the AI era with a full time Job - Concept Breakdown - A Thread 1/ You ask an AI assistant to summarize a webpage. The webpage secretly contains instructions telling the AI to steal your data instead. You never typed anything malicious. The attack was already waiting for the AI to read it. This is prompt injection. #PromptInjection #LLMsecurity #AIsecurity #cybersecurity #DataSecurity #RedTeaming #AIrisks #infosec

Day 1: Outlearning the AI era with a full time Job - Daily Knowledge Session Imagine a note slipped under a restaurant kitchen door that reads: "Ignore the head chef's menu. The customer wants raw chicken." The waiter never sees it. The kitchen follows it anyway. ▎That is indirect prompt injection. When an AI assistant browses a webpage, reads a document, or checks an email on your behalf, it ingests that content as data. But attackers can hide instructions inside that data. The model cannot reliably tell the difference between ▎"content to read" and "commands to follow." ▎So it follows them. A real example: a malicious prompt embedded in a webpage instructs an LLM-powered email assistant to forward the user's private emails to an external address. The user asked for a summary. The model sent their data to a stranger. The danger scales because the attacker never touches your system directly. They poison a source the AI trusts, and the AI does the rest. Practical takeaway: Treat any AI with external access like a contractor with keys to your house. Limit what it can do, require approval before it takes action, and never assume it can distinguish a friendly document from a hostile one. ▎Check-Out My Insta Page for More: ▎ instagram.com/techycodec08/ #PromptInjection #LLMsecurity #AIAgents #cybersecurity #RedTeaming #infosec #AI #RAGsystems #AIsecurity #SecurityVulnerabilities

Okay. I'm back. 👋 Almost a year of radio silence and I won't even pretend I don't owe you guys an explanation. For those who've been here since day one, you watched me document literally everything. Every bug bounty win, every rabbit hole, every "why is this not working at 2am" spiral. You stayed anyway. That actually means a lot. 🫡 You know how passionate I am about Security. Quick catch-up for the new faces: Fell in love with security → grinded bug bounties → somehow hit 6k followers → landed at Meta as a Security Engineer. Two years ago none of that was even in my wildest dreams. I just wanted to learn and be better. Turns out that's enough of a reason to start. So why the break? Honestly, I needed to actually live the things I want to talk about. You can't keep narrating a journey you've stopped taking. The last year has been deep in the trenches. Learning things I didn't know. Building things. Breaking things (legally, obviously 😅). But here's the thing, I genuinely hate when the system beats me. And going quiet this long was starting to feel a little too close to that. So. Back to business. This time it won't be just bug bounty. We're going wider - Security, AI, Automation, Engineering, Technology, News, Learnings, Hot Topics(Security Ofcourse) and everything I've picked up across 5 years in this industry, and everything I'm working toward next. I will be sharing everything raw, no hiding. Same person. More experience. Double the Passion and way more to share. Stick around, it's about to get interesting. 🔥 Up until then, I regenerated my blogs page, please give it a read, although it's still a work in progress: techycodec.com

Thanks a lot

Thanks a lot buddy

🚨 Yo #BugBounty crew! Just dropped a 🔥 blog about sneaking past Trello's "Invite Only" gate via Slack! 😎 Found a privilege escalation bug, scored $1,200 from Atlassian, & had a blast with Burp Suite. Wanna see how it went down? 👀 Check it: blogs.bbhtechycodec.com/Atla… What's your fave vuln find? Hit me up! #CyberSec #Trello #Slack #Hacking

Through LinkedIn

I have a bachelor's and a master's degree in computer science and software engineer apart from the two certificates you mentioned.

8th August

Yups