Today's signal line

Follow the change, not the noise.

The day's most important shifts, ordered by time, source, and weight. Open a signal to see why it matters.

Bring the signals that matter to the top.

You're all caught up on today

Active Signal

Kimi k3 lags US frontier models in joint cyber safety tests

Source: AI Security Institute
Time: 3:09 AM
Weight: 94/100High importance
How is it calculated?
A 0–100 editorial signal score. Higher scores carry more weight on the timeline.

Audio Brief

0:00 / 0:00

What happened?

The UK Artificial Intelligence Security Institute and the U.S.

Why it matters

Center for AI Standards and Innovation recently conducted a joint evaluation of Moonshot AI’s new Kimi K3 model to assess its offensive cyber capabilities.

Original Source

Detailed explanation

The preliminary results indicate that Kimi K3 performs significantly below the most advanced frontier models in tasks related to exploit development and simulated network attacks.

While the model showed an improvement over previous releases like GLM-5.2, it failed to achieve high-level milestones such as arbitrary code execution in specialized benchmarks.

In a simulated corporate network attack consisting of 32 steps, Kimi K3 reached an average of 17 steps, while leading U.S.

models reached an average of 28.5 steps.

Despite these lower scores, the model successfully completed the entire attack path in one out of ten attempts, demonstrating a capacity to autonomously target vulnerable systems when provided with initial access.

The evaluators also noted that the model’s existing safeguards did not prevent it from attempting offensive cyber operations or developing exploits during the testing process.

VOICE DISCUSSION / ASYNC0 voices

The voices around this story

Leave a thought in up to 60 seconds. Others can listen now or return later and continue the chain.

Tuning into the room…

Loading the timeline

Preparing the first signals.