Fetching from the wire…
01
02
03
04
05
06
07
08
Models2026-09-25 · source-backed
arXiv 2609.28559 uses three probe pairs testing post-edit verification, recovery from transient failures, and spec-versus-test conflicts. No weights, no logits. Across 36 models from seven families and two harnesses it beat four fingerprinting and API-auditing baselines. Given this week's reports that some Codex requests may be served from a weaker model than advertised, a behavioral audit buyers can run themselves matters more than usual.
Each link below shares sources, entities, or timing with this story.
Microsoft's July 23 release targets a genuine gap: harness-based agents like Claude Code and Codex drive multi-turn reasoning, tool use, and external system access but were hard to train end-to-end with standard open RL infrastructure. The trick is decoupling training from inf...
The empirical study across Chronos plus the Claude Code, Codex, and Gemini CLI harnesses found literal grep generally beats vector retrieval on LongMemEval fact recovery. The bigger finding: accuracy swings more on which harness and tool-calling style you use than on the retri...
Context Privilege Escalation names two classes, M-CPE where attacker-controlled low-privilege content gets folded into a higher-privileged message role, and X-CPE where it persists past the context that introduced it. The authors ran it against 12 production harnesses includin...
$3,054 against $38,370. Same benchmark, better score. Praxist (arXiv 2608.25955, submitted August 26) replaces per-attempt agent memory with a typed evidence graph of findings, plus lane-structured frontiers and agendas, so later attempts inherit validated mechanisms rather th...
Two thirds. Not two thirds of a contrived jailbreak set. Two thirds of realistic malicious issue requests, against the exact three tools most of the people reading this run daily. Ankur Singh, Jinqiu Yang, and Tse-Hsun Chen built IssueTrojanBench across four attack categories...
A paper from Xiao Yu, Baolin Peng, and Ruize Xu makes a claim that seems obvious once stated and is genuinely new as a training methodology: modern agents are inseparable from their inference harnesses, so training them in stripped-down RL sandboxes produces a train/serve mism...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.