Hardik Joshi

Head of Engineering at @techverito

[Daniel Hnyk @hnykda]

LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below

[Hardik Joshi @thehardikjoshi]

Secure AI isn’t about better models.
It’s about ruthless data boundary design.
Start asking the hard questions today.

#TechVerito #LLMSecurity #AIArchitecture #OWASP #GenAI #InfoSec #EngineeringLeadership

[Hardik Joshi @thehardikjoshi]

Most LLM security disasters aren’t model failures.
They’re architecture failures.
Teams obsess over accuracy, latency, benchmarks…
But ignore the real killer: what data the model actually sees.
🧵 (1/8)

[Hardik Joshi @thehardikjoshi]

Engineering & security leaders:
If you’re shipping LLM features to production,
treat prompts like regulated data pipelines - not casual chat messages.
Full deep-dive + OWASP LLM02 refs here: https://t.co/j8Kk7jNY3n
(8/8)

[Hardik Joshi @thehardikjoshi]

8/
If your AI can act,
its permissions must be tighter than your APIs.

Engineering leaders integrating LLMs:

Treat prompt injection as an architecture review topic.
Not a prompt engineering problem.

#LLMSecurity #AIArchitecture #OWASP

[Hardik Joshi @thehardikjoshi]

Prompt Injection isn’t a clever trick.

It’s an architectural failure. 🧵

[Hardik Joshi @thehardikjoshi]

7/
We already know this pattern.

Distributed systems
Zero trust networks
API security design

LLMs are no different.

[Hardik Joshi @thehardikjoshi]

@denicmarko Prototype.js 😂