digg/ai
2d ago

ExploitGym Benchmarks AI Agents Turning Vulnerabilities Into Real Attacks

——0——
Original post
moyix
Brendan Dolan-Gavitt#843@MOYIX

This paper looks VERY interesting because they managed to get access to run the experiments on models without guardrails

4:33 PM · May 14, 2026 View on X
ORIGINAL POSTBrendan Dolan-Gavitt#843Brendan Dolan-Gavitt@MOYIX

This paper looks VERY interesting because they managed to get access to run the experiments on models without guardrails

11:33 PM · May 14, 2026 · 9.5K Views
REPLYBrendan Dolan-Gavitt#843Brendan Dolan-Gavitt@MOYIX

ArXiv link: https://arxiv.org/abs/2605.11086

Brendan Dolan-GavittBrendan Dolan-Gavitt@moyix

This paper looks VERY interesting because they managed to get access to run the experiments on models without guardrails

11:33 PM · May 14, 2026 · 9.5K Views
11:38 PM · May 14, 2026 · 997 Views
ExploitGym Benchmarks AI Agents Turning Vulnerabilities Into Real Attacks · Digg