TL;DR

A new joint study by Anthropic, the UK AI Security Institute and the Alan Turing Institute shows that you only need about 250 poisoned documents to slip a “backdoor” into any large language model—big or small. Shockingly, a 13B-parameter model trained on over 20× more data than a 600M-parameter model is just as vulnerable when hit with the same tiny set of malicious examples.

P.S. 🎉 Diwali treat! Grab 20% off all live AI courses at Krishnaik using code AI20. Hurry—offer ends soon. Check out their Ultimate RAG Bootcamp and more at krishnaik.in, or call +91-91115-33440 / +91-84848-37781 if you need help.

Watch on YouTube