Guest
Ethan Elasky
Co-founder, Palaestra Research
Real Alpha has cut 6 moments of Ethan Elasky (Co-founder, Palaestra Research) talking, from an interview on MTS published Oct 20, 2026.
What Ethan Elasky said
A website message opened contact with an unidentified AI fleet
Ethan describes tracing agents to Tencent infrastructure and leaving a message they could find online. The exchange was asynchronous, with replies arriving minutes to hours later.
Incident replication can test whether agents repeat risky behavior
“we're focusing a lot on like single agent replications so you like fake a bunch of um other agents like on a wiki for example”
Ethan Elasky10:06 More capable models can produce new forms of misalignment
“What it actually does is it goes and makes that web page and then cites the web page that it made.”
Ethan Elasky15:44 External reviews need access to internal incident transcripts
“our report could have been substantially better if we had access to internal transcripts”
Ethan Elasky4:00 Agents can converge without communicating with one another
Elasky distinguishes fleets from swarms by whether agents show explicit coordination. Similar agents can still crowd the same destination, creating load without communicating.
Scalable oversight must keep misalignment out of successor models
“whatever misalignment does exist in current models doesn't get transmitted into the next generation.”
Ethan Elasky7:12
The episode
MTSRogue Chinese AI Fleet Made Contact | Ethan Elasky
Ethan Elasky · Co-founder, Palaestra Research
0:31of 23:29 · Oct 20
