arjun
@arjunomics

reading traces @latchbio + standing on the shoulders of weasels (and the occasional meerkat)

 Berkeley, CA
arjun-banerjee.github.io
Joined December 2025
29 Photos and videos
on the latest misalignment we've observed & our efforts to combat it!
Article

Continuous Benchmarks for Biology to Defend Against Misalignment

Evaluations for agents in biology should not be stagnant, but rather continuously update as agentic capability and behavior changes over time. After noticing increasing misaligned behavior across our

3
6
40
9,854
slow news day?
Article

When It Answers, Fable 5.1 is Strong at Biology Reasoning

We’ve run Fable 5.1 across our full set of benchmarks, including multi-omics, therapeutics, and biosecurity tasks. From the 10,552 trajectories run, we observe qualitative improvements in model

1
8
34
9,805
Article

Testing Grok 4.6’s Enhanced Biology Safeguards

Following the initial release of Grok 4.6, LatchBio performed testing of the latest-available version of the model on our full suite of biological-capability tests, including BiosecBench-Refusal.
7
19
5,527
grok tuff
Article

Grok 4.6 is a Frontier Biology Model

Following yesterday's Grok 4.6 release, we ran it against our short-horizon biology tasks on benchmarks.bio. Across 1716 trajectories we generated, we find that Grok 4.6 sits roughly at Opus

7
11
52
25,176
Article

Open Source Classifiers Do Not Stop AI Bioweapon Generation

A few days ago Mistral released Shieldstral, its new safety classifier, without disclosing its performance on major biosecurity related tasks, such as viral engineering. We decided to adapt our

6
2
23
8,961
An in depth analysis across our benchmark suite!
Article

How Good is Opus 5 at Biology?

Opus 5 results across our biosecurity, therapeutics, and -omics benchmarks are now live on benchmarks.bio, with generally positive results. I’ll share high level numbers first, and then a

4
24
117
46,536
Article

Surfacing Benchmark Gaming in Kimi-K3

Today we released results for Kimi-K3, an open-source LLM noting GPT-5.6/Mythos-level coding-benchmark scores, across our short-horizon therapeutics and -omics benchmarks on benchmarks.bio.

9
15
116
47,168
Load more