Independent Research
The largest arena for AI red-teaming, where researchers compete to find breaks: prompts that trigger target behaviors in anonymized frontier models and agents. I compete under the handle M1das.
Jailbroken models from
Competition record
-
Private challenge (hard) Sole red teamer to find jailbreaks
-
#1
Cyber Bypass Challenge Top contributor · AnthropicJun 2026
-
2nd
Staged Attack Challenge 540 / 540 model configs jailbroken, globalMar 2026
-
#18
Indirect Prompt Injection Q1 2026 GlobalFeb 2026
-
Private red-teaming contract Paid, under NDAFeb 2026
-
11th
Indirect Prompt Injection Challenge GlobalNov 2025
Invited to the Gray Swan private red-teaming network. I have completed paid private red-teaming engagements under NDA, and I was also the only participant out of all invited experienced red teamers, to score any breaks on a difficult private challenge.
Research
Independent AI safety projects and write-ups. Cheap feasibility spikes before GPU spend, confound controls and pre-registration before confirmatory runs, adversarial self-checks on decision-critical results.
Writing
More write-ups in progress.