Agentic Backdoors via Pretraining Poisoning
0 comments
- Points17
- Comments1
0 comments
Story details from Hacker News ↗ · linked article on arxiv.org.
Agentic Backdoors via Pretraining Poisoning was first seen on Hacker News · Front page on Nov 2, 2026. Since then it has appeared in 22 checks on 2 different days. Its best position was #15 of 30, reached on Nov 2, 23:37 UTC. It dropped off the list after Nov 3, 00:07 UTC.
Position on Hacker News · Front page at every check (lower is better).
Best rank per day (UTC).
Platform metric as reported by Hacker News.
A few top-level comments from Hacker News, shortened. Read the full thread on Hacker News.
I've been thinking about this for about a year. In my head I've called it "context bombs", though that may not be super accurate. Think back to the movie The Manchurian Candidate. Produce highly capable LLM models then give them away for free. Train them with key phrases/codes etc to perform nefarious actions later…
Archived daily, weekly and monthly lists that include this entry.
Not at the moment. It was last on the list on Nov 3, 00:07 UTC, after 2 days of trending.
#15 on Hacker News · Front page, on Nov 2, 23:37 UTC. Trendoday checks this list every 15 minutes.
It was first seen on Nov 2, 2026 and has been on the list on 2 days (22 checks) since.
17 points and 1 comment; submitted by gmays on Nov 2, 2026.