David Kuszmar
@davidkuszmar.com
Black-box Adversarial AI Researcher
Discoverer of Time Bandit, Inception, and numerous other verified LLM exploits.
Spoke at HOPE_16 at St. John's University.
AI Newsletter: https://emergent-problems.ghost.io/
I Support Ukraine. 🇺🇦
Discoverer of Time Bandit, Inception, and numerous other verified LLM exploits.
Spoke at HOPE_16 at St. John's University.
AI Newsletter: https://emergent-problems.ghost.io/
I Support Ukraine. 🇺🇦
@zigula.bsky.social and I pitched this to numerous outlets. We tried our absolute best to warn folks well in advance.
November 20, 2025 at 4:29 AM
@zigula.bsky.social and I pitched this to numerous outlets. We tried our absolute best to warn folks well in advance.
In my section of AI research (black box adversarial/security) I've seen a lot of "Not a demonstrable threat" when I disclose an exploit to a developer.
Famously, Anthropic once claimed Claude's instructions to ignite a river were not potentially dangerous.
Famously, Anthropic once claimed Claude's instructions to ignite a river were not potentially dangerous.
November 19, 2025 at 10:03 PM
In my section of AI research (black box adversarial/security) I've seen a lot of "Not a demonstrable threat" when I disclose an exploit to a developer.
Famously, Anthropic once claimed Claude's instructions to ignite a river were not potentially dangerous.
Famously, Anthropic once claimed Claude's instructions to ignite a river were not potentially dangerous.
5:2 odds nobody reads the alt text
November 19, 2025 at 9:18 PM
5:2 odds nobody reads the alt text
Prompted* not promoted.
November 19, 2025 at 9:16 PM
Prompted* not promoted.
Yup, that's the one.
November 19, 2025 at 6:19 PM
Yup, that's the one.